The Interaction of Silicon Self-Interstitials and Substitutional Carbon inSilicon Based Heterostructures

Malcolm S. Carroll

A DISSERTATIONPRESENTED TO THE FACULTYOF PRINCETON UNIVERSITY

IN CANDIDACY FOR THE DEGREEOF DOCTOR OF PHILOSOPHY

RECOMMENDATION FOR ACCEPTANCEBY THE DEPARTMENT OF

ELECTRICAL ENGINEERING

JUNE 2001

ii

© Copyright 2001 by Malcolm S. Carroll.All rights reserved.

iii

Welcome, O life! I go to encounter for the millionth time the reality of experience and

to forge in the smithy of my soul the uncreated conscience of my race.

J. Joyce, Portrait of an Artist as a Young Man

iv

The Interaction of Silicon Self-Interstitials and Substitutional Carbon in Silicon-Based

Heterostructures

Malcolm S. Carroll

Abstract

The increasing drive to reduce the size of transistors to sub-100 nm dimensions for

overall improved performance is ultimately limited by the control over the dopant pro-

files. Ultra-sharp profiles can be obtained by techniques like ion-implantation or low-

temperature epitaxy, making the primary challenge to control thermal diffusion of

dopants during subsequent fabrication steps after the initial profile is formed. This

work examines two approaches to obtain structures with sharp doping profiles: (1)

reduce the dopant diffusivity through the incorporation of substitutional carbon; and

(2) reduce the thermal budget of the preparation steps for silicon epitaxy creating addi-

tional growth flexibility for novel design of structures with sharp dopant profiles.

This work’s contribution to the first approach include: (i) development of a

new gas chemistry in the Princeton rapid thermal chemical vapor deposition (RTCVD)

reactor for increased substitutional carbon incorporation in low germanium concentra-

tion films, (ii) quantification of the substitutional carbon effect on the dopant diffusiv-

ities in and near SiGeC layers, (iii) discovery and understanding that substitutional

carbon reacts directly with the silicon self-interstitial in a one-for-one “kick-out” like

v

diffusion promotion reaction, (iv) measurement of the interstitial injection rate during

oxidation, (v) demonstration that the carbon-interstitial reaction can effectively “sink”

all excess interstitials introduced by ion-implantation or oxidation. Finally, a low tem-

perature surface cleaning technique for RTCVD epitaxy was also developed as part of

the second approach, which has been used to fabricate ultra-sharp phosphorus profiles.

vi

Acknowledgements

However disproportionately small these acknowledgements are to my gratitude, I hope

nevertheless that the following acknowledgements are appreciated.

Developing understanding is often necessarily a lonely pursuit but for those times that

it wasn’t I wish to express great appreciation for the unguarded and genuinely enjoy-

able exchange of perspective and insight between the students of EMD. I especially

appreciate the cooperative efforts throughout my first year here with Pavel Studenkov,

Jie Yao, Amlan Majumdar, and Xin Wan, the impressions of which go very deep. To

all of Sturm group I am grateful for making a great laboratory atmosphere in which to

work. Special thanks, goes to the RTCVD crowd, Chia-Lin Chang for his introduction

to and help with the reactor and later to Eric Stewart and Haizhou Yin with whom I

have enjoyed countless lively discussions ranging from siphons and gaskets to atomic

hopping and Aharonov-Bohm oscillations. What fun it has been! I will miss it. To all

the EMD faculty I am thankful for providing an excellent and exciting program. I am

furthermore grateful to professors Lyon and Wagner for reviewing this thesis.

I am most deeply thankful to my thesis advisor Jim Sturm, from whom I have learned

the most of all. The combined aggressive optimism, limitless well of insight, and rare

talent to communicate it all has been a remarkable laboratory inspiration. I consider

myself extremely lucky to have had the opportunity to work with such a rare combina-

tion of excellence both as innovator, researcher and teacher.

vii

A special thanks is reserved for Tom Hebner for introducing the basketball hoop to

C424-28 (although office baseball was a really bad idea!) and to Gautam Parthasarathy

and Florian Pschenitzka for making a supportive and entertaining office atmosphere.

Some direct contributors to this work, which can not go without mention: M. Valenti,

L. D. Lanzerotti, T. Büyüklimanli, M. Yang, M. Gokhale, C. Rafferty, H. Rücker, J.

Stangl, E. Napolitani, and D. De Salvador.

Special mention also goes to family and friends all of whom knowingly or not were

giving me most appreciated support. To Irena, I am especially grateful for the serenity

she has brought to my often angst-full soul. To my parents and brother I dedicate this

thesis. My ways, so much a synthesis of yours, this work, so much then yours as mine.

Thank you all.

viii

Contents

Abstract iv

Acknowledgements vi

1. Introduction

1.1 Motivation 1

1.2 Outline 2

2 Growth and Characterization of Pseudomorphic Si1-x-yGexCy

2.1 Introduction 5

2.2 <100> SiGe Layers Pseudomorphically Strained to <100> Silicon 5

2.2.1 SiGe lattice constant 5

2.2.2 SiGe Epitaxy 7

2.3 Carbon Alloyed with Silicon and SiGe 11

2.3.1 Carbon incorporation in Silicon and SiGe 11

2.3.2 Substitutional Carbon in Silicon and Germanium 12

2.4 Measuring Substitutional Carbon 13

2.5 Carbon solid solubility in epitaxially grown Si 15

ix

2.5.1 Carbon solid solubility in epitaxially grown SiGe 18

2.6 Low temperature epitaxy with Disilane 19

2.6.1 Si epitaxy grown by disilane 19

2.6.2 SiGe epitaxy grown by disilane 21

2.6.3 Substitutional Carbon Incorporation in Si and SiGe grown with Disilane 27

2.7 Summary 20

2.8 References 41

3 Reduced Boron Diffusion in the Presence of Substitutional Carbon

3.1 Introduction 43

3.2 Boron Diffusion in Silicon 44

3.2.1 Boron diffusion mediated by Si interstitials 44

3.2.2 Oxidation enhanced diffusion 52

3.2.3 Transient enhanced diffusion 54

3.3 Boron diffusion in SiGeC 55

3.3.1 Introduction of Cs to reduce boron diffusion 56

3.3.2 HBT Process Conditions 57

3.3.3 Extraction of diffusion constant (method) 60

3.3.4 Boron diffusivity found in SiGe and SiGeC HBT bases 61

3.4 Summary 67

3.5 References 69

x

4 Substitutional Carbon as a Silicon Self-Interstitial Sink

4.1 Introduction 71

4.2 Previous reports on reduced boron diffusivity in the presence of carbon 72

4.3 Phosphorus diffusion above the SiGeC layer 83

4.4 Carbon profile after annealing 88

4.5 Summary 90

4.6 References 92

5 Quantitative Measurement of the Surface Silicon Interstitial Boundary Condi-

tion and Silicon Interstitial Injection into Silicon during Oxidation

5.1 Introduction 94

5.2 Interstitial profile above buried SiGeC layers 95

5.2.1 Test structures 95

5.2.2 Extraction of Diffusivity Enhancement and Interstitial Profile 100

5. 3 Boron Diffusion in a Linear Diffusivity Gradient 105

5.4 Interstitial Injection Rate 107

5.4.1 Calculation of Interstitial Injection Rate 107

5.4.2 Discussion and Comparison to Other Work 111

5.5 Summary 115

5.6 References 116

xi

6 Diffusion Enhanced Carbon Loss from SiGeC Layers due to Oxidation

6.1 Introduction 118

6.2 Substitutional carbon loss after oxidation 119

6.3 Quantification of carbon loss 128

6.4 Double carbon layer experiment 133

6.5 Simulations 138

6.5.1 Model of carbon diffusion 138

6.5.2 Modelling of carbon diffusion profiles 142

6.6 Summary 152

6.7 References 154

7 Low-Temperature Preparation of Oxygen- and Carbon-Free Silicon and Sili-

con-Germanium Surfaces for Silicon and Silicon-Germanium Epitaxial Growth

by Rapid Thermal Chemical Vapor Deposition

7.1 Introduction 156

7.2. Standard Growth and Characterization Procedures 158

7.2.1 Cleaning and Growth Procedures 158

7.3 Characterization of Interfaces 160

7.4. Ex-situ wet cleaning 162

7.4.1 Introduction 162

xii

7.5. Contamination from Reactor, Load-Lock, and Laboratory Environment 168

7.5.1 Experiment 168

7.5.2 Discussion 174

7.6. In-situ Wafer Cleaning 174

7.6.1 200-400°C pre-bakes 174

7.6.2 700-800°C Bakes in Hydrogen 176

7.6.2.1 700°C Bakes in Hydrogen 177

7.6.2.2 800°C Bake in Hydrogen 180

7.6.3. Discussion 184

7.7. Summary 190

7.8 References 192

8 Conclusion

8.1 Summary 194

8.2 Future Work 195

Appendix A 200

Appendix B 213

Publications and Presentations Resulting from this Thesis 213

Journal Articles 213

xiii

Refereed Conference Papers 214

Conference Presentations 215

xvi

Figure 2.1 Lattice constant of silicon, relaxed SiGe and strained SiGe 7Figure 2.2 Schematic of the Princeton RTCVD 9Figure 2.3 Growth rates of Si and SiGe 11Figure 2.4 Substitutional C incorporation vs. temperature and partial pressure 18Figure 2.5 O, C, Ge profiles of Si0.93Ge0.07 layer grown using disilane 23

Figure 2.6 Ge concentration vs. germane (0.8% in hydrogen) flow rate 24Figure 2.7 XRD rocking curve of Si0.93-xGe0.07Cx layers vs. carbon content 25

Figure 2.8 PL spectra of Si0.93Ge0.07 layer 26

Figure 2.9 O, C, Ge profiles of (a) Si0.926Ge0.07C0.004 layer and (b) Si0.996C0.004 28

Figure 2.10 Total C vs. methylsilane flow for both SiC and SiGeC 29Figure 2.11 Growth rates for SiC and SiGeC grown w/ disilane 30Figure 2.12 XRD rocking curve of Si0.996C0.004 layer 32

Figure 2.13 Substitutional carbon vs. total carbon 33Figure 2.14 FTIR absorbance spectra 35Figure 2.15 Integrated absorbance vs. integrated substitutional carbon 37

Figure 3.1 Schematic representation of diffusion 46Figure 3.2 OED of boron and phosphorus in silicon 47Figure 3.3 Energy level diagram for interstitial-boron diffusion 48Figure 3.4 Boron profiles in silicon before and after annealing 53Figure 3.5. Schematic diagram of SiGe HBT structure 56Figure 3.6 Schematic conduction band diagram reduction of collector current 57Figure 3.7 HBT Gummel plots and collector current vs. base-collector voltage 58Figure 3.8 B, C, Ge profiles of SiGe and SiGeC HBT’s after implant 60Figure 3.9 HBT saturation currents extracted from Gummel plots 62Figure 3.10 B diffusivities from fitted collector saturation currents 64Figure 3.11 Calculated Early voltages for adjusted excess interstitials 66

Figure 4.1 Test structure to study boron or phosphorus OED/TED 74Figure 4.2 B profiles in pure Si of as-grown and annealed (897°C for 30 min) 76Figure 4.3 B profiles from sample w/ SiGeC layer before and after annealing 78Figure 4.4 Diffusivity enhancement vs. carbon in SiGe(C) 80Figure 4.5 Schematic diagram of the interstitial concentration for three cases 82Figure 4.6 TED dependence on carbon levels in SiGe(C) barrier layer 83Figure 4.7 P profiles of before and after annealing in nitrogen 85Figure 4.8 P profiles before and after annealing in oxygen ambient 86Figure 4.9 Relative P diffusivity vs. carbon levels in buried Si(GeC) layers 88Figure 4.10 C profiles before and after annealing 90

xvii

Figure 5.1. Schematic diagram of the test structures 96Figure 5.2. B and C profiles before and after annealing (30 minutes at 850°C) 97Figure 5.3. B and C profiles before and after annealing (120 minutes at 850°C) 98Figure 5.4. B and C profiles before and after annealing (240 minutes at 750°C) 99Figure 5.5. B and C profiles before and after annealing (960 minutes at 750°C) 100Figure 5.6. Boron diffusivity enhancements for all samples and depths (750°C) 102Figure 5.7. Boron diffusivity enhancements for all samples and depths (850°C) 103Figure 5.8. Interstitial super-saturation at the surface 105Figure 5.9. Simulation of boron diffusion profile in self-interstitial gradient 107Figure 5.10. Total interstitial injected into the silicon 110Figure 5.11. Comparison to loop-defect measurement 114

Figure 6.1 Schematic cross section of structures 120Figure 6.2 C profiles from samples of structure A 121Figure 6.3 C profiles from samples of structure B 122Figure 6.4 XRD rocking curves of structure B before and after annealing at 850°C 123Figure 6.5 Total carbon in SiGeC layer (SIMS & XRD) vs. time (structure B) 125Figure 6.6 Summary of total carbon lost from the SiGeC layers 129Figure 6.7 Summary of oxidation enhanced carbon loss from the SiGeC layers 130Figure 6.8 Schematic diagram of the kick-out recycling mechanism 133Figure 6.9 Schematic cross section of the double SiGeC layer structure 134Figure 6.10 B and C profiles from the double buried SiGeC layer and B diffusivity135Figure 6.11 C profiles from the double buried SiGeC layer and simulation 144Figure 6.12 Simulated substitutional, interstitial C and interstitial silicon profiles 145Figure 6.13 Simulated defect currents 149

Figure 7.1. Typical PL spectrum at 77 K of a Si0.8Ge0.2 161

Figure 7.2. Relative SiGe/Si (PL) intensity vs. HF pH used 166Figure 7.3. C, O and F detected at Si/SiGe interfaces 167Figure 7.4. O, C and Ge profiles at 700°C hydrogen bake interface 170Figure 7.5. O, C and Ge profiles at SiGe/Si test interfaces 172Figure 7.6. Integrated O and C vs. different conditions w/out high T cleaning step173Figure 7.7. SiGe/Si PL from SiGe/Si after 300ºC bake for different times 176Figure 7.8. SiGe/Si PL from SiGe/Si layers after 1 min. 800ºC bakes vs. pressure 181Figure 7.9. C and O levels at SiGe/Si interfaces after 1 min., 800ºC bake vs. P 183Figure 7.10. C and O level vs. time at 800ºC 184Figure 7.11. O, C and P after interrupt growth to form ultra-sharp P layer 186

xviii

Figure 7.12. Schematic diagram of contributions in RTCVD reactor to O desorption/absorption 188

Figure 8.1. Boron diffusivity vs. germanium content 197

1

Chapter 1

Introduction

1.1 Motivation

The scaling of transistors to ever smaller sizes has created an increasing demand for

control over dopant profiles in silicon devices. Production of source-drain profiles

long ago moved away from solid-source in-diffusion techniques to relying on ion

implantation to form tailor made dopant profiles for devices. As device dimensions

have shrunk control of the dopant diffusion length has become ever increasingly criti-

cal in defining the minimum size of a device. Past generations of dopant profile engi-

neering have relied on shrinking the thermal budgets of the device fabrication process

to reduce the thermal broadening of profiles enough to remain within the limits of the

design parameters for junction depth and gate-lengths. Smaller sizes have, therefore,

traditionally meant lower temperatures, however, processing steps like ion implanta-

tion, which create point defects like the silicon self-interstitial that promote diffusion

of common dopants like boron and phosphorus, limit the success of this approach. Ion

implantation is known to produce enough excess silicon self-interstitials to enhance

boron and phosphorus diffusivities thousands of times above their intrinsic diffusivi-

ties, making this effect the dominant source of diffusion at low temperature.

One novel approach to control boron diffusion has been the incorporation of car-

bon in Si or SiGe, which dramatically reduces the boron diffusivity and has been

shown to suppress the otherwise enhanced diffusivities resulting from processing steps

2

like ion-implantation. For this reason, the incorporation of carbon has been used to

significantly improve the thermal stability of some important device structures, e.g. the

SiGeC heterojunction bipolar transistor (HBT). However, neither the mechanism

responsible for the reduced diffusivity nor a quantitative understanding of the carbon

effect are well established. In this work, the local and non-local effects of the presence

of substitutional carbon on the boron diffusivity in Si and SiGe is quantitatively exam-

ined, the thesis of which is that the substitutional carbon reacts directly with the silicon

self-interstitial in a simple one-to-one reaction (“kick-out”), depleting the local inter-

stitial concentration and thereby reducing the local boron or phosphorus diffusivities.

Also during this thesis work, a low temperature cleaning step for RTCVD epitaxy was

developed for the fabrication of novel device structures. This step is useful for device

structures that require the growth sequence to be interrupted so that epitaxial layer may

be processed and returned to the reactor for a subsequent growth of a second epitaxial

layer.

1.2 Outline

In chapter 2, growth of thin Si1-x-yGexCy epitaxial alloy layers by rapid thermal chem-

ical vapor deposition (RTCVD) at Princeton is reviewed. This chapter includes a dis-

cussion of new work on the introduction of the gas (disilane) into the Princeton reactor

for the successful incorporation of high substitutional carbon concentrations ( ~ 0.5%

C) in Si and low germanium fraction SiGe (7%). In chapter 3, the technological moti-

3

vation for RTCVD grown device structures with sharp dopant profiles is illustrated by

demonstrating the increased thermal stability of the SiGeC HBT over the SiGe HBT.

It is found that a typical high performance design using a thin (~ 20 nm) highly boron

doped base (~1020 cm-3) is very sensitive to small boron diffusion lengths ~ 1-2 nm

that significantly reduce the electrical performance of the transistor, and increased

control over the dopant profiles in the HBT structure is established by incorporating

0.5% substitutional carbon in the base. In chapter 4, the enhanced boron diffusion

below SiGeC layers, resulting from excess interstitials injected by oxidation or ion-

implantation, is found to depend on the amount of substitutional carbon in the SiGeC

layer. The dependence of the boron enhanced diffusion on substitutional carbon con-

centration introduces the question of how many substitutional carbon are necessary to

react with a single self-interstitial? Using boron marker layers to map out the silicon

self-interstitial profile under different oxidation conditions above the SiGeC layer, in

chapter 5, the interstitial surface boundary condition is determined and the number of

injected interstitials during oxidation is quantified. In chapter 6, using the injection

rates determined in the previous chapter, a single injected interstitial is found to

remove one substitutional carbon from the SiGeC layer in the simplest case consistent

with a simple interstitial “kick-out” mechanism. Finally an indirectly related work is

discussed in chapter 7 on the production of sharp dopant profiles by an interrupt and

regrowth technique, which depends critically on a low thermal budget cleaning step.

Finally in chapter 8 a summary of the results is combined with a short discussion of

4

some suggestions for future directions of research in this area.

5

Chapter 2

Growth and Characterization of Pseudomorphic Si1-x-yGexCy

2.1 Introduction

In the following chapters the reaction of substitutional carbon in silicon and

SiGe with crystal point defects will be discussed. That work required the growth of

test structures containing, pseudomorphically strained SiGeC and SiC layers, and

required the characterization of the total substitutional carbon content in those alloy

layers. This chapter, therefore, reviews epitaxial growth of pseudomorphically

strained SiGe to silicon by rapid thermal chemical vapor deposition (RTCVD) and

then discusses how the substitutional carbon concentration in SiC and SiGeC is mea-

sured in order to describe growth and characterization of SiC and SiGeC alloys at

Princeton.

2.2 <100> SiGe Layers Pseudomorphically Strained to <100> Silicon

2.2.1 SiGe lattice constant

Silicon and germanium both crystallize in the diamond lattice structure, and the two

elements are completely miscible forming a random alloy Si1-xGex with a lattice con-

stant that can vary between aSi (0.5431 nm) and aGe (0.5675 nm) [1], and a bandgap

6

that likewise varies between the silicon and germanium bandgaps of roughly 1.1 and

0.67 eV (room temperature), respectively [2, 3]. The two elements are isoelectronic

and the alloy has been very useful providing device designers a way to engineer the

band-gap of devices for improved transistor performance [4]. However, because there

exists a lattice constant mismatch between the SiGe alloy and silicon, the SiGe alloy

layer must be strained to the silicon lattice (Si is the widely available substrate) to

avoid undesirable misfit defects. If the strained layer is too thick it is energetically

favorable for the strained layer to relax even at the expense of forming interface

defects. Therefore, defect-free strained SiGe layers are constrained to a maximum

critical thickness [5].

The SiGe layer commensurate to the silicon substrate is compressively strained

in the plane parallel to the surface of the silicon, which leads to an increase of the ver-

tical Si1-xGex lattice constant, , in the direction perpendicular to the silicon sub-

strate surface. The vertical lattice constant of the resulting pseudomorphically strained

tetragonal crystal can be described using the elastic theory,

(2.1)

where C11(x) and C12(x) are the elastic constants of Si1-xGex and can be linearly inter-

polated between Si and Ge [5], and the relaxed lattice constant is approximated as the

weighted average Å [1, 5]. The strained and relaxed

⊥SiGe

a

−⋅++⋅=⊥

Si

SiSiGeSiSiGe a

axa

xC

xCxCaxa

)(

)(

)(2)(1)(

11

1211

xxaxaSiSiGe

⋅+−⋅= 225.0)1()(

7

Si1-xGex lattice constants are plotted versus germanium concentration in Fig. 2.1. The

germanium fraction of a pseudomorphically strained Si1-xGex crystal layer with uni-

form composition may, thereby, be deduced from the vertical lattice constant of the

layer using x-ray diffraction techniques.

Figure 2.1 Lattice constant of silicon, relaxed SiGe and strained SiGe as a function ofgermanium fraction.

2.2.2 SiGe Epitaxy

On a regular basis, high quality, uniform, defect free thin films of <100> SiGe are

grown on <100> silicon substrates in the Princeton RTCVD [6, 7] reactor. All test

structures presented in the following chapters were grown in the Princeton reactor. A

8

schematic, Fig 2.2, shows the essential elements of the system. A single 4” wafer is

introduced into an evacuated quartz chamber on a quartz stand through a load-lock.

The load-lock is critical to maintain low oxygen backgrounds in the chamber, neces-

sary to produce silicon with low oxygen concentrations (1018 cm-3 or less) [8]. During

growth the wafer and stand are exposed to a constant flow of H2, 3 slpm, while main-

taining the chamber pressure at 6 or 10 torr. Source gases such as dichlorosilane

(SiCl2H2) are mixed into the H2 to supply the silicon. The crystal growth is initiated

by thermal decomposition of the source gases on the silicon wafer surface. The silicon

wafer is heated radiatively by a bank of twelve 6 kW tungsten-halogen lamps located

below the quartz tube capable of delivering a maximum power of 72 kW. A gold

reflector assembly is situated around the quartz tube to more efficiently couple light

into the wafer.

9

Figure 2.2 Schematic of the Princeton rapid thermal chemical vapor deposition cham-ber.

All gases injected into the reactor are exhausted by a rotary vane mechanical

pump. The absorption of infrared light by silicon is very temperature sensitive due to

the temperature dependence of the intrinsic carrier concentration and the band gap

edge. Therefore, the transmission of infrared light relative to its transmission at room

temperature is a measurement of the wafer temperature. Modulated infrared laser light

(1.3 and 1.55 µm) is passed through the wafer during growth and detected below the

wafer before and during wafer heating to determine the wafer temperature. Because

the growth rate and material composition are strongly temperature dependent the tem-

perature of the wafer must be controlled accurately within a few degrees of the desired

Lamps

WaferDetector

Lock-inAmplifier

Quartz Tube

Gas Inlet

Reflector AssemblyLaser

Laser FiberLoad-Lock

Modulator

10

temperature. Throughout the growth the temperature is maintained with a feedback

loop using the supplied lamp power and wafer absorption as the input and output sig-

nals, respectively [9].

Growth of the SiGe alloy is initiated by introduction of germane (0.8% GeH4

in hydrogen) while growing silicon into the reactor atmosphere, which then decom-

poses with the silicon source gas resulting in a fraction of silicon lattice sites replaced

with germanium. Likewise the silicon or SiGe may be doped n or p type by introduc-

ing phosphine (PH3) or diborane (B2H6) to supply the phosphorus or boron, respec-

tively. Typical growth conditions for this reactor include flows of 3 slpm H2 and 26

sccm dichlorosilane (DCS) holding the reactor pressure at 6 or 10 torr. The resulting

growth rates for these conditions at temperatures between 675-800ºC are shown in Fig

2.3.

11

Figure 2.3 Growth rates of single crystal silicon grown using dichlorosilane at 6 torrflowing 26 sccm and 3 slpm of hydrogen with or without the addition of germane(0.8% in hydrogen) compared to the growth rate of poly-silicon grown using 3 slpm ofhydrogen and 50 sccm of disilane (10% in hydrogen) at 10 torr for temperaturesbetween 550 and 1000°C.

2.3 Carbon Alloyed with Silicon and SiGe

2.3.1 Carbon incorporation in Silicon and SiGe

Because carbon is isoelectronic to both germanium and silicon and also may crystal-

lize in a diamond structure, carbon is also another available element to alloy with sili-

con and germanium, which provides more material flexibility to the device designer.

Furthermore, the Si1-xCx random alloy is very appealing to device designers, because

there is a conduction band offset between silicon and Si1-xCx, making it ideal for elec-

1

10

100

1000

104

8 9 10 11 12 13 14 15

Gro

wth

Rat

e[Å

/min

]

1/kT [eV]

625 5508001000

Si2H6

Si2Cl2H2 + 300 sccm GeH4

Si2Cl2H2 + 100 sccm GeH4

Si2Cl2H2

12

tron-based devices [10], which is absent in the system using SiGe strained commensu-

rately to silicon. Because the lattice constant of carbon crystallized in the diamond

structure is considerably smaller than that of silicon (aC is 0.3546 nm), carbon incor-

poration in SiGe to form SiGeC is also of great interest for its potential to compensate

the compressive strain in SiGe layers grown commensurate to silicon [11] and there-

fore relax the critical thickness constraint on SiGe layers.

2.3.2 Substitutional Carbon in Silicon and Germanium

Below the solid solubility concentration in silicon or germanium carbon occupies sub-

stitutional sites and is relatively benign as substitutional carbon, Cs. The bulk solid

solubility of carbon is high in neither silicon (3-4x1017 cm-3 near the melting point

[12]) nor germanium (108-1010 cm-3 [13]). No stable Ge-C phases are known above

the solid solubility limit of germanium [14]. In silicon, silicon-carbide (SiC) is the

only thermally stable phase. The precipitation of silicon-carbide in silicon with carbon

concentrations far above the solid solubility is, however, known to be extremely slow

without the presence of other catalytic elements such as oxygen [15]. It has been pro-

posed that the slow precipitation of SiC is due to the unusually high surface energies

required to form the silicon carbide precipitate in silicon (8000 erg cm-2, 15 times that

of SiO2 precipitates) [15]. Above solid solubility in silicon or germanium, carbon has

been identified in various meta-stable states including substitutional carbon, intersti-

13

tial carbon Ci, and several carbon interstitial complexes X-Ci including Bs-Ci, and Cs-

Ci [12]. These different states of carbon in silicon are all undesirable, except for sub-

stitutional carbon, because of their effects on the electrical properties of the material.

2.4 Measuring Substitutional Carbon

Total carbon concentrations in silicon and SiGe may be measured by secondary ion

mass spectrometry (SIMS). SIMS is an often utilized tool for determining the chemi-

cal composition of semiconductors and is sensitive to carbon concentrations as low as

5x1016 cm-3. However, no information about the carbon state, i.e. whether it is substi-

tutional, is obtained by this measurement. Uncertainty in SIMS obtained carbon con-

centrations is relatively high and is often as great as 20% [16], leading to significant

difficulties in quantifying the fraction of carbon that is substitutional.

Many states of carbon in silicon are identifiable by their local vibration modes

and their number in silicon are quantifiable by the strength of the absorption line.

There are a large number of carbon local vibration modes, some of which are: intersti-

tial carbon at 922 and 932 cm-1; Cs-Ci (A-Mode) at 594, 722, and 872 cm-1; Cs-Ci (B-

Mode) at 842, 730, 7819 cm-1 (G-Line); Ci-I (silicon interstitial) 959 and 966 cm-1;

Ci-Oi at 865 and 1115 cm-1 (the two most intense); Ci-Oi + I at 940 and 1024 cm-1 and

silicon-carbide at 796 and 960 cm-1. Substitutional carbon is identifiable by its local

vibrational mode at 607 cm-1 or 531 cm-1 in pure silicon or germanium, respectively.

14

In pure silicon the oscillator strength of the Cs mode is known and the concentration of

carbon in silicon can, therefore, be measured. Using Beer’s law, the peak absorption

coefficient of the 605 cm-1 line with a full-width-half-max of 6 cm-1 at room tempera-

ture has been correlated to the substitutional carbon concentration as:

(2.2)

where a is measured in units of cm-1 with an uncertainty of ~ 20% in the proportional-

ity constant [17]. Because there is interference with both substitutional carbon in the

substrate on which the alloy layers are grown and also with a nearby absorption line,

610 cm-1, due to the emission of two phonons in the silicon lattice [18], a careful pro-

cedure of subtracting a substrate IR spectrum (identical to the sample substrate) from

the sample IR spectrum is required to remove the interfering components.

Quantification of substitutional carbon in layers with low total carbon areal

densities (less than 1-2x1015 cm-2) requires a more sensitive indirect technique to

determine the amount of substitutional carbon, due to the uncertainties inherent in the

IR absorption technique. X-ray diffraction is an extremely sensitive method for deter-

mining the lattice constant of crystals. Because the lattice constant of silicon and Si1-

xGex shrink due to the incorporation of the smaller carbon atoms on substitutional sites

in the crystal, the total amount of substitutional carbon may be deduced by the lattice

constant of the SiGeC alloy layer, in a similar way as described in section 2.2.1 for

strained SiGe, by using a Ge to C strain compensation ratio n (i.e. if 1 carbon atom

α××= −217101.1][ cmCs

15

compensates the strain produced by 10 Ge atoms then n = 10). This method requires

that the germanium concentration be determined by an independent measurement.

Values of n range from 8-12 in the literature [19-21] due to uncertainties about the

relaxed lattice constant of SiGeC and the validity of linearly interpolating the elastic

constants in eq’n. 2.1, however, reasonable success has been reported with n = 12 [21],

which is used in this work. Therefore, if a (004) Bragg reflection from a SiGeC layer

with 20% Ge is located at the same angle as that expected from a SiGe layer with 8%

Ge, then we conclude that the SiGeC layer contains 1% Cs.

Substitutional carbon content in Si1-xCx can likewise be deduced from the ten-

sile strained lattice constant as was described in section 2.2.1. The tensile strained lat-

tice constant is predicted by the elastic theory as:

(2.3)

where C11(x) and C12(x) again are the elastic constants of Si1-xCx and is the

relaxed lattice constant of carbon. All of these constants are linearly interpolated

between silicon and diamond (i.e. using Vegard’s law).

2.5 Carbon solid solubility in epitaxially grown Si

As discussed in previous sections, carbon has a very low solubility in silicon and ger-

manium, and the only thermally stable form of carbon in silicon, above the solid-solu-

−⋅⋅++⋅=⊥

Si

SiSiCSiSiC a

axa

xC

xCxCaxa

)(

)(

)(2)(1)(

11

1211

)(xaSiC

16

bility (3x1017cm-3 at the melting point), is silicon carbide (SiC). This makes it

difficult to alloy an appreciable amount of carbon on substitutional sites in either sili-

con or SiGe at thermal equilibrium. It takes, therefore, non-equilibrium crystal growth

circumstances to alloy high concentrations of carbon on substitutional sites of silicon

and SiGe. Kinetically dominated crystal growth, like that in the case of rapid thermal

chemical vapor deposition or molecular beam epitaxy, indeed, has demonstrated the

ability to incorporate up to 2.5% substitutional carbon in silicon and SiGe [22, 23].

The remarkably high substitutional carbon concentrations reported in epitaxi-

ally grown materials has been attributed to a higher solubility of carbon on the surface

of silicon than in the silicon bulk. The surface solubility has been predicted to be as

much as 104 times greater than that in the silicon bulk [24]. However, the fraction of

carbon that is grown on substitutional sites has been found to be very sensitive to the

growth conditions. Using molecular beam epitaxy (MBE), Osten et al. demonstrated

that the fraction of substitutional carbon depended on two key parameters: temperature

and growth rate. By increasing the growth rate, while maintaining the same material

composition they showed that the fraction of substitutional carbon increased with

increasing growth rate for a given temperature. Furthermore, an increase in substitu-

tional carbon fraction was also observed as the temperature was decreased, while

maintaining the same growth rate and material composition. The substitutional frac-

tion dependence on growth rate and temperature was modeled as a competition

between an energy activated time for the carbon species to move from a substitutional

17

surface site to a defect site and the time taken to grow a monolayer of the alloy, after

which the buried carbon would be immobilized on the substitutional site [25]. The

time for defect formation on the surface was proposed to be much faster than in the

bulk, where the substitutional carbon is bound on the substitutional site more strongly

by the additional surrounding atoms.

Initial attempts to grow Si1-xCx by CVD using dichlorosilane and methylsilane

for the silicon and carbon, respectively, were relatively unsuccessful incorporating at

best 20% of the total carbon on substitutional sites [26]. However, growth of Si1-xCx

alloys containing 100% substitutional carbon were demonstrated by Mitchell et al.,

when using silane and methylsilane as precursors for silicon and carbon, respectively

[26]. They reported that it was necessary to use low growth temperatures or high

silane partial pressures to achieve high substitutional carbon fractions, Fig. 2.4. These

results are consistent with Osten’s work and suggest that high growth rates and low

growth temperatures are indeed critical for high Cs incorporation.

18

Figure 2.4 Substitutional carbon incorporation as a function of growth temperatureand silane partial pressure [26].

2.5.1 Carbon solid solubility in epitaxially grown SiGe

Incorporation of substitutional carbon in SiGe has been demonstrated using CVD or

MBE and carbon concentrations as high as 2.5% have been reported in the Princeton

RTCVD reactor using dichlorosilane, germane, and methylsilane as source gases for

silicon, germanium and carbon respectively [27]. The ability to incorporate relatively

high carbon concentrations in CVD epitaxially grown SiGe films, despite using

dichlorosilane gas chemistry (slow silicon growth rate) and reports of lower solid sol-

ubility in pure germanium [13] and SiGe [27], is presumably because of the catalyti-

19

cally enhanced growth rates (Fig 2.3) of the SiGe epitaxial films when germane is

added to the gas chemistry [28]. However, incorporation of high concentrations of

substitutional carbon in pure silicon and SiGe alloys with low germanium concentra-

tion until this work has remained undemonstrated at Princeton. Because it is well

established that the silicon growth rate is significantly higher when using disilane for

the silicon source [29, 30] and for this work it was necessary to grow Si1-xCx and Si1-

x-yGexCy with low germanium fraction we explore the introduction of disilane for

higher substitutional carbon fraction in low germanium fraction alloys in the following

sections.

2.6 Low temperature epitaxy with Disilane

2.6.1 Si grown by disilane

Silicon dioxide, 85 nm thick, was grown on <100> double side polished p-type silicon

wafers, over which poly silicon was grown in the Princeton RTCVD reactor at temper-

atures between 550-800ºC using disilane as a source gas. The oxidized wafers were

inserted into the reactor after oxidation without additional cleaning steps. All epitaxy

using disilane was done using a growth pressure of 10 torr while flowing 3 slpm of H2

and 50 sccm of 10% disilane 90% hydrogen mixture. Growth rates for the poly-silicon

(Fig 2.3) were found using the thickness after growth, determined by fitting the optical

reflectance spectra of the poly-silicon on oxide using the Nanospec Automated Film

Thickness System (Model No. 4100), and dividing by the deposition time. Because

20

the growth rate varies across the wafer surface due to temperature variation, all thick-

ness measurements were taken within a 1 cm radius of the center of the substrate,

where the temperature is measured during growth. Growth rates may be slightly

higher than reported because they are calculated using the total growth time, which

does not account for any poly-silicon nucleation time on the oxide. However, nucle-

ation times reported for disilane-grown silicon on oxide for similar temperatures in

other CVD systems represent less than 5% of the total growth times for each tempera-

ture examined in this experiment [31]. The polycrystalline nature of the deposited sil-

icon was confirmed by the observation of a reflectance peak at 276 nm, which is not

present in amorphous silicon [32]. All UV reflectance measurements were also done

with the 4100 Nanospectrometer.

Growth rates of single crystal silicon or SiGe grown at 625ºC using a flow rate

of 26 sccm of dichlorosilane (3 slpm H2, 6 torr) and germane flow rates of 0-300 sccm

are compared to the growth rate of poly-silicon using disilane (Fig. 2.3). The growth

rate of poly-silicon displays a strong temperature dependence increasing rapidly from

30 to 253 Å/min. for temperatures of 550ºC and 625ºC, respectively, which is charac-

teristic of a reaction-limited growth. At higher growth temperature, 700-800ºC, the

growth rate rolls off and is not as temperature sensitive indicative of mass-transport

limited growth. The growth rates of poly-silicon at lower temperatures, 550ºC to

625ºC, are much higher than those obtained with dichlorosilane and they are compara-

ble to or higher than the growth rate of SiGe grown using dichlorosilane and germane,

21

in which substitutional carbon concentrations as high as 2.5% were achieved. If the

critical conditions for high substitutional carbon incorporation are primarily high

growth rates and low growth temperatures, then this is very promising for the growth

of Si1-xCx. Indeed, in the following sections 100% substitutional carbon incorporation

to form Si1-xCx and low Ge (7%) content SiGeC alloys are demonstrated.

2.6.2 SiGe epitaxy grown by disilane

Pseudomorphically-strained (100) SiGe layers were grown using disilane and germane

and examined using secondary ion mass spectrometry, photoluminescence and X-ray

diffraction. Float-zone <100> silicon substrates polished on both sides were cleaned

using a standard ex-situ and high temperature in-situ clean. Growth commenced with

a thick (~0.25 µm) silicon buffer layer grown at 6 torr flowing 26 sccm dichlorosilane

and 3 slpm of H2 at 1000ºC. After the buffer layer was grown and the dichlorosilane

was switched off, the wafer was allowed to cool for 10 minutes. The hydrogen flow

rate was maintained constant throughout the entire growth sequence even during cool-

ing. As discussed on page 9, the temperature of the wafer is monitored by comparing

the infrared transmission at the growth temperature to that at room temperature.

Because the room temperature transmission is sensitive to both changes in thickness

and surface roughness it may change appreciably due to the growth of the silicon

buffer layer. The growth was interrupted, therefore, at this time to remeasure the room

temperature infrared transmission through the silicon wafer to improve the accuracy of

22

the temperature control. The wafer was then reheated to 625ºC and the reactor pres-

sure was adjusted to 10 torr, after which 50 sccm of disilane (10% Si2H6 in hydrogen)

was injected growing silicon for 30 seconds (~15 nm) followed by the SiGe layer initi-

ated by germane injection (27 sccm of 0.8% germane in H2 in this case). The SiGe

layers were then capped with a silicon layer grown by turning off the germane flow at

625ºC for 15 seconds then turning off the disilane flow, increasing the temperature to

800ºC and injecting dichlorosilane for 10 minutes maintaining the reactor pressure at

10 torr. SIMS of the SiGe layer (Fig. 2.5) show a low oxygen and carbon back-

grounds, very sharp junctions and uniform germanium concentrations throughout the

layer. The germanium concentration

23

Figure 2.5 Oxygen, carbon and germanium profile obtained by SIMS of aSi0.93Ge0.07 layer grown using disilane (sample 2800). The buffer layer and cap weregrown using dichlorosilane.

dependence on germane flow rate for layers grown at 625ºC is compared to the con-

centration dependence observed in films grown with dichlorosilane at 625ºC in Fig

2.6. The germanium incorporation is higher in the films grown using dichlorosilane

for the same germane flows, which may be in part due to the smaller growth rate of the

SiGe films grown using dichlorosilane. The growth rate for Si0.93Ge0.07 at 625ºC was

faster than polysilicon grown with disilane, 314 Å/min and 253 Å/min, respectively.

The faster growth rate of SiGe compared to silicon is consistent with reports of cata-

1016

1017

1018

1019

1020

1021

1022

0 200 400 600 800 1000

Ato

mic

Con

cent

rati

on[c

m-3

]

Depth [nm]

Ge

O

C

24

lytic increased growth rates when germane is added to disilane [29] or dichlorosilane

[28].

Figure 2.6 Germanium concentration of SiGe layers grown at 625°C using 3 slpm ofhydrogen and dichlorosilane (26 sccm) or disilane (50 sccm, 10% in hydrogen) at 6 or10 torr respectively as a function of germane (0.8% in hydrogen) flow rate. Germa-nium concentrations were determined by X-ray diffraction. Note: growth rates varywith germane flow rate.

X-ray diffraction rocking curves around the silicon 004 Bragg reflection show

two peaks, Fig 2.7. The narrow satellite peak centered around -700 arcseconds is the

Bragg reflection from the larger lattice constant of the crystalline SiGe layer. The

peak position corresponds to a lattice constant expected from a 100% strained layer

0

8

16

24

32

40

0 100 200 300 400 500

Ger

man

ium

Con

cent

rati

on[%

]

Germane Flow [sccm]

Si2H6

Si2Cl2H2

25

Figure 2.7 X-ray diffraction rocking curve around the (004) Bragg reflection of<100> Si0.93-xGe0.07Cx layers grown on <100> silicon with carbon content varyingfrom x=0-0.0081 measured in Princeton. Note: the total carbon content is indicatedabove each of the corresponding Si0.93-xGe0.07Cx Bragg reflections (samples 2800,2798, 2827 with increasing carbon respectively).

with 6.88% Ge agreeing well within the uncertainty of the SIMS measurement. Fur-

thermore, photoluminescence measurements of the silicon capped Si0.93Ge0.07 layer at

77 K using an argon-ion laser as an excitation source (Fig 2.8) show two bright lumi-

nescence peaks corresponding to the transverse optical (TO) phonon replicas from the

silicon substrate and the SiGe layer. The separation in energy between the silicon and

SiGe TO replicas is a measure of the difference of energy bandgaps between the two

100

1000

104

105

106

-1000 -500 0 500

Inte

nsit

y[a

rb.u

nits

]

Delta Theta [Arcsec]

Si substrate

0.0%

C

0.4 4

%C

0.81

%C

26

materials, ~ 56 meV. The bandgap of strained SiGe can be roughly estimated as that of

silicon reduced by ~7-8 meV/ %Ge, which is consistent with XRD and SIMS mea-

surements of the germanium concentration and the bright luminescence from the layer

indicates the high quality defect free nature of the layer [6]. Note that the no-phonon

line usually observed from strained SiGe layers, located 58 meV greater than the SiGe

TO line (the energy of the TO phonon), is much less intense in low germanium frac-

tion SiGe [33] and is obscured by the interference with the silicon TO line.

Figure 2.8 Photoluminescence (PL) spectra of a Si0.93Ge0.07 layer (Sample #2800)capped with ~ 45 nm of silicon. The PL was done at 77 K and the excitation power

was 1.5 W/cm2.

0

5

10

15

900 1000 1100 1200

Inte

nsit

y[a

rb.u

nits

]

Energy [meV]

SiGe TO Si TO77 K

27

2.6.3 Substitutional Carbon Incorporation in Si and SiGe grown with Disilane

Thick layers of Si1-xCx and Si1-x-yGexCy were grown using disilane, methylsilane and

germane as source gases and examined using XRD, SIMS, and PL. In the case of the

Si1-x-yGexCy layers, the exact same procedures were followed as described in the pre-

vious section for the growth of the Si1-xGex layers except that while growing the Si1-

xGex layers at 625ºC different flows rates (0.09-1 sccm) of methylsilane were intro-

duced into the gas stream. In the case of Si1-xCx, the growth procedure was identical

to that described in the previous section except that germane is never introduced to the

gas stream, thinner silicon caps were grown (17 minutes, 700ºC, grown with dichlo-

rosilane), and after the silicon cap was finished the wafer was annealed in-situ at

800ºC for 5 minutes with a constant flow of 3 slpm of H2. It has been proposed, as

discussed earlier, that carbon can incorporate on interstitial sites as well as substitu-

tional sites. This final thermal anneal is done to drive any interstitial carbon out of the

alloy layer and corresponds to an interstitial carbon diffusion length of ~ 1 mm [12],

which is much greater than the alloy layer thickness of approximately 100-300 nm.

Substitutional carbon has been shown to be thermally stable in silicon and SiGe for

short (i.e. 5 minute) anneals at 800ºC [25], so no appreciable loss of substitutional car-

bon is expected due to this thermal anneal. The concentration profiles of a Si1-x-

yGexCy and a Si1-xCx layer are shown in Fig 2.9 (a) and (b), respectively. High carbon

incorporation is achieved in the SiGe and silicon layers while maintaining a very simi-

28

lar germanium fraction and a relatively low oxygen background. The dependence of

total carbon incorporation, measured using SIMS, on methylsilane flow rate for all

SiGeC and SiC samples grown using disilane is shown in Fig 2.10.

Figure 2.9 Oxygen, carbon and germanium profile obtained by SIMS of (a)Si0.926Ge0.07C0.004 layer and (b) Si0.996C0.004 layer grown using disilane (samples2798 and 2963, respectively). The alloy layers were grown at 625°C at 10 torr using50 sccm of disilane (10% in hydrogen), and 3 slpm of hydrogen.

No significant germanium fraction dependence was observed on carbon con-

centration. However, increases in oxygen concentration have been observed in all lay-

ers with increased carbon concentration, and those layers with the highest carbon

concentrations also tend to have the highest oxygen backgrounds. The cause for the

increased oxygen concentrations in the carbon layers is not known at this time and

could be due to contamination in the methylsilane or perhaps is caused by surface

1017

1018

1019

1020

1021

1022

0 200 400 600 800 1000

Ato

mic

Con

cent

rati

on[c

m-3

]

Depth [nm]

1017

1018

1019

1020

1021

0 200 400 600 800A

tom

icC

once

ntra

tion

[cm

-3]

Depth [nm]

(a) (b)

Ge

O

C

O

C

29

chemistry involving carbon that is inherently more susceptible to oxygen incorpora-

tion.

Figure 2.10 Total carbon concentrations determined by SIMS are displayed as a func-tion of methylsilane flow rate for both SiC and SiGeC samples. Note: the methylsi-lane flow rates are plotted as total methylsilane content in the gas stream (notincluding the hydrogen content flowed at the same time)

Growth rates for SiGeC and SiC were also determined from SIMS measure-

ments and growth times (Fig 2.11). The methylsilane flow rate has very little influ-

ence on the growth rate of the SiGeC layer. However, the growth rate of SiC with the

higher methylsilane flow rate (167 Å/min) is ~ 27 % less than that grown with the

lower flow rate (230 Å/min). Layer thickness can vary as much as 30% over the cen-

0

0.5

1

1.5

0 0.4 0.8 1.2

SiCSiGeCT

otal

Car

bon

Con

cent

rati

on[%

]

Methysilane Flow [sccm]

625°C625°C, 10 torr

30

ter of the wafer surface due to temperature variation, therefore, it can not be concluded

from just these two measurements whether the growth rate depends on the methylsi-

lane flow rate.

Figure 2.11 Growth rates determined from SIMS and layer growth times as a functionof methylsilane flow rate for SiC and SiGeC samples. Note: the methylsilane flowrates are plotted as total methylsilane content in the gas stream (not including thehydrogen content flowed at the same time).

The X-ray diffraction rocking curves from the Si0.93-xGe0.07Cx structures (x =

0.0044 or 0.0081) are overlaid on the rocking curve obtained from the Si0.93Ge0.07

structure and a clear positive peak shift due to the strain compensation of the substitu-

tional carbon is observed, Fig 2.7. Furthermore, the sample with 0.81% total carbon

0

50

100

150

200

250

300

350

0 0.2 0.4 0.6 0.8 1 1.2

Gro

wth

Rat

e[Å

/min

]

Methysilane Flow Rate [sccm]

625°C, 10 torr

SiGeC

SiC

31

has a lattice constant smaller than that of silicon (indicated by the positive shift from

the silicon substrate Bragg reflection), demonstrating that the amount of substitutional

carbon is more than enough to fully compensate the compressive strain from the 7%

Ge in the layer. The intensity of photoluminescence of the TO phonon replica from

the SiGeC layers at 77 K is slightly less intense than that from the SiGe layer consis-

tent with previous reports of reduced carrier lifetimes due to carbon incorporation in

SiGe [25, 34]. The x-ray rocking curve of the Si0.996C0.004 structure (measured by

SIMS) has a Bragg reflection from the SiC layer 500 arcsec from the silicon Bragg

reflection (Fig 2.12). The magnitude of angular shift in this case also corresponds to

~100% substitutional carbon. No photoluminescence at 77 K was observed from this

layer, however, because the bandgap offset (~ 25 meV, 68 meV / %C [11]) is small it is

unclear whether this is because of poor carrier confinement, high recombination due to

carbon defects or perhaps a combination of both. The total carbon measured by SIMS

32

Figure 2.12 X-ray diffraction rocking curve around the (004) Bragg reflection of<100> Si0.996C0.004 layers grown on <100> silicon (sample 2963). The Si0.996C0.004layer was grown at 625°C.

for all SiGeC and SiC samples grown using disilane are plotted against the substitu-

tional carbon determined by XRD measurements and is summarized in Fig 2.13. The

amount of substitutional carbon concentration measured in the alloy layer by XRD

depends on the choice of relaxed lattice and elastic constants that are used in the linear

interpolation (Vegard’s law) for eq’n. 2.3. It is not known at this time what the appro-

priate choice of material should be for the necessary constants, e.g. diamond or sili-

con-carbide [21] or even whether Vegard’s law is appropriate [35, 36]. Broad error

bars on the substitutional carbon axis indicate the spread of possible values that can be

10

100

1000

104

105

-500 0 500 1000

Inte

nsit

y[c

ount

s/se

c]

Delta Theta [arcsec]

Si substrate

0.4%

C

33

obtained using the range of material constants between diamond and silicon-carbide.

Within the uncertainty of this measurement 100% substitutional carbon incorporation

for carbon concentrations up to 0.81 % are possible.

Figure 2.13 Substitutional carbon determined from Bragg reflection angular shifts(XRD) plotted against the total carbon measured in the sample using secondary ionmass spectrometry (SIMS). The solid line indicates the substitutional carbon concen-tration corresponding to 100% substitutional carbon in the layer. A 20% error range(due to uncertainty in SIMS) is indicated for total carbon and the range of uncertaintyshown for the substitutional carbon represents different possible values that may beobtained based on Vegard’s Law.

Normal incidence FTIR absorption measurements were made on 1 cm2 sam-

ples of as-grown wafers at room temperature following the procedures outlined in the

ASTM standard [17]. A typical absorbance spectra of infrared light transmitted

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1 1.2Subs

tiut

iona

lCar

bon

Con

cent

rati

on[%

]

Total Carbon Concentration [%]

SiC

SiGeC

100%

Subs

titut

iona

l

34

through a float-zone silicon sample 495 µm thick is displayed in Fig. 2.14 (a). The

absorption peak centered at 610 cm-1 is due to the two-phonon emission in the silicon

substrate. The phonon-line interferes with the substitutional carbon local mode

absorption at 605 cm-1 and dominates the absorption signal. As discussed in section

2.4, it is necessary to subtract the two-phonon absorption from the total absorption

spectra to quantitatively measure the substitutional carbon concentration in the alloy

layer grown on the substrate. This can be done by measuring the absorption spectra of

an identical silicon substrate and subtracting it from the absorption spectra of the alloy

layer and substrate. To illustrate the sensitivity of this method to differences in the

substrate thickness, the absorption of the 495 µm thick float-zone substrate is sub-

tracted from a different 505 µm float-zone substrate. A difference of 10 microns

between two different silicon float-zone substrates produces an observable contribu-

tion to the absorption signal (Fig. 2.14 (a)) and would introduce an error in the quanti-

fication of the carbon concentration. Special care, therefore, is taken to use the

identical substrate for subtraction. The identical substrate for subtraction is prepared

by etching the epitaxial layers off the top and bottom surface of a part of the same

wafer with the epitaxially grown alloy layer (a micron or less on each side). The lay-

ers were wet-etched with a solution of HF:NH3:H2O (1:9:10). The reflectivity is uni-

form between 500-1000 cm-1 after etching and is similar to the original polished

substrates. The resulting difference in thickness resulted in no appreciable two-

35

phonon emission or carbon signals (Fig. 2.14 (b) labeled background). A typical

absorption spectrum of a Si1-xCx sample (Fig 2.14 (b)) after subtracting its identical

substrate spectra, shows no signs of carbon in defect states (location of the defect lines

are indicated on the figure (see section 2.4 for references), while exhibiting a strong

absorption line at 605 cm-1 corresponding to the substitutional carbon local mode at

room temperature.

Figure 2.14 (a) Above, the FTIR absorbance spectra of a pure FZ silicon 495 µmthick substrate (subtraction of only the surrounding nitrogen ambient). Below, thespectra of the substrate from above after subtracting the spectra of a second pure sili-con FZ wafer 505 µm thick. (b) Above, the FTIR absorbance spectra of a 300 nm thickSi0.996C0.004 layer (sample 2963) grown on a FZ silicon substrate after subtracting theabsorption spectra of its own substrate after wet-etching the epitaxially grownSi0.996C0.004 layer off. Below, the absorbance spectra of a Si control sample after sub-tracting the absorption spectra of the same silicon sample after undergoing the samewet etch used to remove the epi-layers from the other samples.

(b)(a)

0

0. 2

0. 4

0. 6

0. 8

500 600 700 800 900 1000Wavenumber [cm -1]

Abs

orba

nce

[cm

-1]

2-phononC

s

Background

Si

-0.025

0

0.025

0.05

0.075

500 600 700 800 900 1000

Abs

orba

nce

[cm

-1]

Wavenumber [cm -1]

Cs

ß-Si

C

Ci

Ci-I

Ci-O

i

Cs-

Ci

Backgroundx10

36

The integrated absorption measured in each of the SiGeC (squares) and SiC

(circles) layers is plotted against the integrated substitutional carbon. The integrated

substitutional carbon is calculated from the substitutional carbon concentration deter-

mined by X-ray diffraction (Fig. 2.13) multiplied by the thickness of the layer (deter-

mined by SIMS) and includes the contribution from both the layers grown on top and

bottom surfaces of the substrate (Fig. 2.15). The measured absorption is compared to

an estimate of that expected based on the absorption strength of the substitutional car-

bon local mode from the literature (solid line in Fig. 2.15) [17]. The relatively good

agreement and the lack of signal from other common carbon defect local modes fur-

ther supports the assertion that the carbon incorporated into the alloy layers using disi-

lane and methylsilane gas chemistry is substitutional carbon.

37

Figure 2.15 Integrated absorbance of the 605 cm-1 substitutional carbon local vibra-tion mode as a function of integrated substitutional carbon deduced from XRD mea-surements. The solid line indicates the expected integrated absorbance from theliterature [17].

The maximum substitutional carbon concentration attainable with these

growth conditions remains unexamined. The highest carbon incorporation attained in

the Princeton reactor was 2.5% grown at 575°C flowing 400 sccm germane [23, 34].

This resulted in a growth rate of ~25 Å/min, which is slightly less than the expected

silicon growth rate using disilane. Supposing that the amount of substitutional carbon

is limited only by growth rate and temperature of the growth, Si1-xCx layers with

~2.5% appear attainable from the Princeton reactor. Furthermore, incorporation of

0

0.2

0.4

0.6

0.8

1

0 5 1015 1 1016 1.5 1016

Inte

grat

edA

bsor

banc

e[c

m-2

]

Integrated Carbon [cm-2]

SiC

SiGeC

ASTMEsti

mate

38

substitutional carbon in silicon may be more efficient than in SiGe grown at the same

temperature and growth rate because germanium is suggested to reduce the carbon sol-

ubility [13, 27]. For this work, however, such high carbon concentrations were unnec-

essary. Si1-xCx alloys with high substitutional carbon concentrations have been

reported grown using disilane in ultra high vacuum chemical vapor deposition systems

but this is the first report, to the authors knowledge, of Si1-xCx alloys grown using dis-

ilane in a low pressure (LP)CVD system. A table summarizing all the measurements

of all disilane grown layers discussed in this chapter is included for the reader’s refer-

ence (Table 2.1).

39

Table

2.1Sum

mary

ofSam

plesG

rown

usingD

isilane

2961

5

0.9

0

625

0.5

-

-

17.0

-

-

-

-

2824

5

0.44

0

625

1

-

-

10.4

-

-

-

85

2826

5

0.22

0

625

5.5

-

-

7.2

-

-

-

58

2800

5

0.22

0

625

7

220 ± 10

6.5

7.3

-

-

-

57.5

2801

5

0.22

0.09

625

7

210 ± 15

6.4

7.0

8.5x1019

5.04x1019

0.235

52.5

2825

5

0.22

0.19

625

5.5

195 ± 10

6.8

7.2

1.9x1020

1.82x1020

0.266

x

2798

5

0.22

0.21

625

7

210 ± 15

7

6.9

2.15x1020

1.62x1020

0.454

45

2827

5

0.22

0.45

625

5.5

200 ± 15

6.8

7.2

4.05x1020

3.44x1020

0.762

x

2964

5

-

0.19

550

60

150 ± 8

-

-

1x1020

9.08x1019

0.281

x

2799

5

-

0.19

625

3

55 ± 8

-

-

2x1020

-

-

x

2963

5

-

0.19

625

11

305 ± 15

-

-

2.2x1020

1.92x1020

0.827

x

2765

5

-

1.0

625

3

50 ± 5

-

-

7x1020

-

-

x

Sample

Si2H6 Flow [sccm]

GeH4 Flow [sccm]

SiCH6 Flow [sccm]

Temp. (set-point)[Celsius] ± 10 ºC

Time [min] ± 0.5

Layer Thickness[nm]

Ge (SIMS)[atomic %] ± 2 %

Ge (XRD)[atomic %] ± 1 %

C (SIMS)[cm-3] ± 20 %

C (XRD) [cm-3]± 5x1019 cm-3

IntegratedFTIR absorb. [cm-2]

± 0.04

PL (ESi- ESiGe)TO[meV] ± 1 meV

40

Note: The difference between location of the TO peaks (photon energy) coming fromthe SiGe(C) layer and the Si bulk is listed in the PL column. X indicates that the mea-surement was taken but no signal was obtained from the sample and - indicates that nomeasurement was taken. The volume of pure gas flow is listed rather than the totalmixture, which includes hydrogen.

2.7 Summary

Growth and properties of <100> SiC and SiGeC commensurately strained to <100>

silicon were discussed including the method of measuring substitutional and total car-

bon in the alloy layers. The growth of SiC required the introduction of disilane to the

Princeton reactor, which provided the appropriate gas chemistry to incorporate rela-

tively high concentrations of substitutional carbon in silicon and germanium poor

SiGeC layers. The work on disilane is believed to be the first demonstration of disi-

lane grown SiC with low pressure chemical vapor deposition. In the following chapter

one technological advantage of carbon incorporation in the SiGe(C) base layer of the

heterojunction bipolar transistor (HBT), the effect of substitutional carbon on boron

diffusion, will be discussed. This motivates the discussion in later chapters of the rela-

tionship between substitutional carbon and boron diffusion.

41

2.8 References

[1] J. Dismukes, L. Ekstrom, and R. Paff, J. Phys. Chem., vol. 68, pp. 3021, 1964.[2] D. Lang, R. People, J. Bean, and A. Seargent, Appl. Phys. Lett., vol. 47, pp.

1333, 1985.[3] R. Braunstein, A. Moore, and F. Herman, Phys. Rev., vol. 109, pp. 695, 1958.[4] H. Kroemer, Proc. IEEE, vol. 70, pp. 13-25, 1982.[5] C. Van de Walle and R. Martin, Phys. Rev. B, vol. 34, pp. 5621, 1986.[6] J. C. Sturm, P. V. Schwartz, E. J. Prinz, and H. Manoharan, J. Vac. Sci. Tech. B,

vol. 9, pp. 2011, 1991.[7] J. C. Sturm, H. Manoharan, L. Lenchyshyn, M. Thewalt, N. Rowell, J. P. Noel,

and D. Houghton, Phys. Rev. Lett., vol. 66, pp. 1362, 1991.[8] P. V. Schwartz and J. C. Sturm, J. Electrochem. Soc., vol. 141, pp. 1284, 1990.[9] J. C. Sturm, P. M. Garone, and P. V. Schwartz, J. Appl. Phys., vol. 69, pp. 542,

1990.[10] W. Faschinger, S. Zerlauth, G. Bauer, and L. Palmetshofer, Appl. Phys. Lett.,

vol. 67, pp. 3933, 1995.[11] O. G. Schmidt and K. Eberl, Phys. Rev. Lett., vol. 80, pp. 3396, 1998.[12] G. Davies and R. C. Newman, “Carbon in monocrystalline Silicon,” in Hand-

book on Semiconductors, T. S. Moss, Ed., 1994, pp. 1558.[13] R. I. Scace and G. A. Slack, J. Chem. Phys., vol. 30, pp. 1551, 1959.[14] B.-K. Yang, M. Krishnamurthy, and W. H. Weber, “Incorporation and stability

of carbon during low-temperature epitaxial growth of Ge1-xCx (x<0.1) alloys

on Si(100): Microstructural and Raman studies,” J. Appl. Phys., vol. 82, pp.3287, 1997.

[15] W. J. Taylor, W. Y. Tan, and U. Goesele, Appl. Phys. Lett., vol. 62, pp. 2336,1995.

[16] R. G. Wilson, F. A. Stevie, and C. W. Magee, “Test method for substitutionalcarbon atoms of silicon by infrared absorptions,” in ASTM Book of Standards,1989, pp. 252 (F123-86).

[17] ASTM, “Test method for substitutional carbon atoms of silicon by infraredabsorption,” in ASTM Book of Standards, vol. F123-86, 1986.

[18] F. Johnson, “Lattice absorption bands in silicon,” Proc. Phys. Soc. London,vol. 73, pp. 265, 1959.

[19] S. C. Jain, H. J. Osten, B. Dietrich, and H. Ruecker, Semicond. Sci. Technol.,vol. 10, pp. 1289, 1995.

[20] J. L. Hoyt, T. O. Mitchell, K. Rim, V. Singh, and J. F. Gibbons, Thin SolidFilms, vol. Aug., 1997.

42

[21] D. De Salvador, M. Petrovich, M. Berti, F. Romanato, E. Napolitani, A. Drigo,J. Stangl, S. Zerlauth, M. Muehlberger, F. Schaeffler, G. Bauer, and P. C.Kelires, Phys. Rev. B, vol. 61, pp. 13005, 2000.

[22] K. Brunner, K. Eberl, and W. Winter, Phys. Rev. Lett., vol. 76, pp. 303, 1996.[23] C.-L. Chang, A. St. Amour, and J. C. Sturm, Appl. Phys. Lett., vol. 70, pp.

1557, 1997.[24] J. Tersoff, Phys. Rev. Lett., vol. 74, pp. 5080, 1995.[25] J. H. Osten, Thin Solid Films, vol. 367, pp. 101-111, 2000.[26] T. O. Mitchell, J. L. Hoyt, and J. F. Gibbons, Appl. Phys. Lett., vol. 71, pp.

1688, 1997.[27] J. P. Liu and H. J. Osten, Appl. Phys. Lett., vol. 76, pp. 3546, 2000.[28] P. M. Garone, J. C. Sturm, P. V. Schwartz, S. A. Schwartz, and B. J. Wilkens,

Appl. Phys. Lett., vol. 56, pp. 1275, 1990.[29] T. I. Kamins and D. J. Meyer, Appl. Phys. Lett., vol. 61, pp. 90, 1992.[30] C. Li, S. John, and E. Quinones, J. Vac. Sci. Technol. A, vol. 14, pp. 170, 1996.[31] Y. Z. Hu, D. J. Diehl, C. Y. Zhao, C. L. Wang, Q. Liu, A. Irene, K. N. Chris-

tensen, D. Venable, and D. M. Maher, J. Vac. Sci. Technol. B, vol. 14, pp. 744,1996.

[32] K. Pangal, “Hydrogen-Plasma-Enhanced Crystallization of HydrogenatedAmorphous Silicon Films: Fundamental Mechanisms and Applications,” inElectrical Engineering. Princeton: Princeton University, 1999.

[33] V. Venkataraman, “Physics and Applications of Si/SiGe Modulation DopedStructures,” in Electrical Engineering. Princeton: Princeton University, 1994.

[34] C.-L. Chang, “Properties and Applications of Crystalline SiGeC alloys,” inElectrical Engineering. Princeton: Princeton University, 1998.

[35] J. Martins and A. Zunger, Phys. Rev. Lett., vol. 56, pp. 1400, 1986.[36] P. C. Kelires, Phys. Rev. B., vol. 75, pp. 8785, 1997.

43

Chapter 3

Reduced Boron Diffusion in the Presence of Substitutional Carbon

3.1 Introduction

In the previous chapter the growth and characterization of Si1-x-yGexCy was described.

In this chapter boron diffusion and the effect of substitutional carbon on boron diffu-

sion in the base of the Si1-x-yGexCy heterojunction bipolar transistor (HBT) will be

examined. This chapter includes a discussion of the microscopic model of boron dif-

fusion in silicon, i.e. the role of silicon interstitials in solid-state diffusion of boron in

silicon, and a quantitative examination of the effect of incorporation of substitutional

carbon on the boron diffusivity in Si1-x-yGexCy. This highlights the technological

importance of controlling boron diffusion in the SiGe(C) HBTs. The introduction of

substitutional carbon is found to significantly reduce the boron diffusivity improving

the performance of the HBT, which motivates later chapters examining the detailed

reaction between the substitutional carbon atoms and silicon interstitial. The work in

this chapter was published largely in reference [1].

44

3.2 Boron Diffusion in Silicon

3.2.1 Boron diffusion mediated by Si interstitials

The random thermal motion of particles is described by diffusion theory. Diffusion

has been extensively studied in various different states of matter, i.e. gas, liquid and in

solids. Although diffusion theory is over a hundred years old and diffusion of boron in

silicon has been technologically important for many decades, the detailed mechanisms

that produce boron diffusion are still not completely understood. The description of

boron diffusion that follows [2] is a predictive theory that adequately agrees with

experiment in a vast regime of annealing conditions and is therefore used in many

common process simulators, e.g. TSUPREM4 (commercially offered by Avant) or

PROPHET [3].

Boron is known to dissolve onto the substitutional lattice sites of silicon when

the concentration is below its solid solubility, and is designated as boron in the substi-

tutional state Bs. The boron atom is relatively strongly bound in this state and is con-

sidered immobile, therefore the diffusivity of the Bs is for all intents and purposes

zero. Early experiments using electron paramagnetic resonance (EPR) identified a

boron atom that shares the same site as a silicon atom, named an Interstitials, which is

mobile [4, 5]. It has been assumed, then, that the observed boron diffusion is mediated

by the mobile boron defect state, Bi, that has non-zero diffusivity. The mobile boron

defect state is formed when a silicon point defect, i.e. silicon vacancy or interstitial

atom, is captured by the substitutional boron atom.

45

The observed intrinsic diffusivity results, therefore, from the Bs reaction with

the intrinsic population of free silicon point defects, leading to the random thermal

motion of the mobile Bi defects. The point defect responsible for boron diffusion in

the temperature range of 700ºC < T < 900ºC, which is considered in this work, has

been experimentally determined as the silicon interstitial atom in various charge states

[6, 7] (Fig. 3.1). The chemical reaction leading to the negatively charged defect is

expressed as:

(3.1)

where is the ionized substitutional boron atom and I is the silicon interstitial. It

has also been experimentally determined that the boron diffusivity is linearly depen-

dent on the hole concentration at hole concentrations above the intrinsic concentration.

This has led to the proposal of an overall neutral mobile defect formed between a pos-

itively charged interstitial and a negatively charged ionized dopant [8-10]:

(3.2)

where p is a hole, is a silicon interstitial with a hole localized on it, and is the

resulting neutral mobile boron defect.

−− →+is

BIB

−s

B

0iss

BIBpIB →+→++ +−−

+I 0i

B

46

Figure 3.1 Schematic representation of (a) an interstitial silicon atom near a substitu-tional boron atom in a silicon lattice (for diagrammatic simplicity a square lattice isused) and (b) the resulting mobile boron interstitial(cy) formed by the binding of theinterstitial silicon atom to a boron.

The temperature dependence of the intrinsic boron diffusivity is described well

over a large temperature range by a thermal activation energy of ~3.4 eV [2], plotted in

Fig. 3.2 (b). The activation energy describes the combined formation energy of the

mobile defect and the boron defect’s migration energy. The energy of formation of an

BI

B

I reacts with

substitutionalboron

Boron interstitialcy(Si and B share site)is highly mobile

Si Interstitial ( I )

(a) (b)

47

Figure 3.2 (a) diffusion enhancement ratios (Dox / Dintrinsic) during oxidation forboron and phosphorus [7, 20, 31, 34-36], and (b) the boron and phosphorus diffusivi-ties either during annealing in inert (intrinsic) or oxygen ambient [2, 7, 20, 31, 34-36].

interstitial defect is in the neighborhood of 3.7 eV [2, 11]. The binding energy of the

silicon interstitial by the substitutional boron in the charge neutral state has been esti-

mated to be 0.9 eV [2, 4, 12], which agrees relatively well with the experimentally

determined activation energy of the total boron diffusivity of 0.6 eV, measured by

electron paramagnetic resonance (EPR) [4] (Fig 3.3). However, the formation and

binding energies of the silicon self-interstitial remains speculative [2, 12] and illus-

trates the lack of fundamental knowledge about the reaction mechanisms that lead to

the boron diffusivity that still remain.

1

10

100

0.7 0.75 0.8 0.85 0.9 0.95 1

Ural [P]Carroll [B]Kuo [B]Roth, Plummer [B]S. Pindl [B]Packan [P]Dunham [P]

OE

DE

nhan

cem

ent

Fac

tor

1000/T [Kelvin-1]

850 750°C10001100°C 900 800

10 -20

10 -19

10 -18

10 -17

10 -16

10 -15

10 -14

10 -13

0.7 0.8 0.9 1

B (No OED)Carroll [B]Kuo [B]Roth, Plummer [B]Pindl [B]P (No OED)Ural [P]Packan [P]Dunham [P]

Dif

fusi

vity

[cm

2s-1

]

1000/T [Kelvin-1]

850 750°C10001100°C 900 800

(a) (b)

0.7 eV -3.4 eV

-2.7 eV

PP

B

B

48

Figure 3.3 Energy level diagram for the dominant interstitial-boron species in borondiffusion. Note: neither the energy of formation for the free silicon interstitial or thebinding energy of the silicon interstitial to the boron substitutional atom have beenmeasured directly.

The reactions in equations 3.1 and 3.2 are assumed to occur rapidly relative to

the diffusion time scale, which means that all species are in thermal equilibrium with

one another at all times during diffusion:

(3.3)

(3.4)

where k0,1 are thermally activated equilibrium constants that describes the detailed

balance between the species. The total flux of boron atoms must include the diffusive

flux of both charged and neutral interstitial boron species and the drift flux of the

charged mobile defect in an electric field (e.g. within the depletion region of a p-n

junction). The combined flux of all species is, therefore [2, 3, 9, 13]:

3.7 eV3.4 eV

0.9 eV0.6 eV

EBs+I

EBi

EBimigration

EBs+SiEBs+Si

EBs+I

EBi

EBimigration

IBkBsi

⋅⋅= −−1

1

pIBkBsi

⋅⋅⋅= −0

0

49

where is the total boron flux, is either the mobile boron defect in the neutral

(n=0) or negatively charged state (n=1), the flux of either the neutral or charged

mobile boron defect, is the diffusivity of one of the two species, n is the charge of

the mobile defect, q is the charge of a single electron, kB is the Boltzmann constant, T

is the temperature and is the electric field. The last term of the flux describes the

drift diffusion due to any built in fields from extrinsic doping conditions, which relies

upon the Einstein relationship to express the mobility of the species as .

Combining eq’n. 3.3, 3.4, and 3.5, the boron flux under intrinsic conditions can

be rewritten as:

(3.6)

where the total boron flux in intrinsic silicon, , from the mobile boron defect spe-

cies is equated to Fick’s 1st law of diffusion for the substitutional boron concentration,

and is the experimentally observed intrinsic boron diffusivity. The remaining

variables are, , the intrinsic interstitial concentration, and the intrinsic free car-

∑∑=

−−

=

⋅⋅⋅−⋅∂∂−==

−

−−

1,01,0

)()(n B

Bni

niB

nBB E

Tk

qDBnqBD

xJJ

ni

ni

ni

r

(3.5)

BJ n

iB−

niB

J −

niB

D −

Er

Tk

qD

B

BD

ni

niB

−

−=µ

x

BIpkDIkDB

xD

x

BDJ s

BB

ni

nB

sBB

iin

i ∂∂⋅+−=

∂∂⋅−=

∂∂= −−

−

=∑ )( **

0*

11,0

**01

*B

J

*B

D

*I*p

50

rier concentration more frequently designated as ni ( is used to avoid confusion in

the notation). The observed diffusivity of the substitutional boron, , can be

equated to the mobile defect diffusivities, the equilibrium constants, and the intrinsic

interstitial concentration from inspection of eq’n. 3.6,

(3.7).

The expression for the intrinsic boron diffusivity highlights its linear dependence on

the silicon interstitial concentration. In cases when the interstitial concentration is per-

turbed from its intrinsic value the observed diffusivity increases proportionally to the

relative interstitial concentration:

(3.8)

where I is the local interstitial concentration. The relationship between the relative

interstitial concentration and the observed extrinsic diffusivity is somewhat different.

Starting from eq’n. 3.6 and assuming that there is no spatial gradient in the interstitial

concentration the boron flux may be expressed as:

where,

(3.10)

*p

*B

D

** sBB

DI

ID ⋅=

⋅⋅⋅+

∂∂××−= −−

=

−∑ E

Tk

qBnqB

xI

IDJ

Bss

n

nBB

r

)(1,0

*(3.9)

*0*

00

0 p

pDpIkDD

BsBBi

⋅==

**0

*1

*01

IpkDIkDDii BBB

+= −

51

(3.11)

and are the observed substitutional boron diffusivities in the negatively

charged and neutral states. All other variables are as previously defined [2, 3, 13].

This relationship is widely used in the extrinsic case to measure the local relative inter-

stitial concentration by comparing the observed diffusivity to the extrinsic diffusivity

expected with no excess interstitials. The extrinsic diffusivity is calculated using the

local hole concentration and the literature values for and . Determination of

the relative interstitial concentration from a boron profile is complicated by the con-

centration dependent diffusivity. The relative interstitial concentration, therefore, is

usually found by numerically fitting an experimentally obtained boron profile to simu-

lated profiles with a constant extrinsic diffusivity multiplied by a single fitting vari-

able, i.e. the relative interstitial concentration. It should be noted that the validity of

eq’n 3.9 relies on an assumption that the diffusivity of the mobile defect and the inter-

stitial concentrations are spatially uniform. If either of these vary spatially, there will

be additional contributions to the flux. Except where otherwise noted, it is assumed in

this work that the additional boron flux due to interstitial or diffusivity gradients is

negligible.

1*1

11

−− == − BsBBDIkDD

i

0Bs

D 1−Bs

D

0Bs

D 1−Bs

D

52

3.2.2 Oxidation enhanced diffusion

Certain silicon fabrication steps result in an increase in the interstitial concentration

above the intrinsic concentration. Increases of silicon interstitials in the silicon lattice

have been identified, for example, by the growth of stacking faults or dislocation loops

below the silicon surface after oxidation [14, 15], which grow by the addition of extra

silicon atoms. Presumably, because of the large mismatch in molecular volume of sil-

icon dioxide and silicon, silicon interstitials are formed at the interface to relieve the

stresses created by the oxidation reaction [16-18]. The silicon interstitials created at

the surface consequently migrate to the stacking faults where they are trapped and

contribute to the growth of the extended defect.

Boron diffusion is also known to increase during oxidation and the oxidation

enhanced boron diffusion (OED) is attributed to the interaction between the substitu-

tional boron and the excess interstitials injected during oxidation. In typical OED

experiments it is found that the boron diffusivity is relatively uniform under the oxi-

dized surface (Fig. 3.4), which is attributed to the relatively uniform interstitial con-

centrations due to long diffusion lengths of the silicon interstitial compared to the

boron diffusion length for the same anneal conditions [2, 19]. Boron and phosphorus

OED have been examined extensively over a wide range of temperatures (Fig. 3.2 (a)),

and the similar diffusivity enhancements are because phosphorus and boron both dif-

fuse primarily by an interstitial mechanism [6, 7].

53

Figure 3.4 (a) boron concentration profiles in pure silicon before and after annealingat 750°C for 240 minutes in either nitrogen or oxygen ambient and (b) the depthdependence of the diffusivity enhancement during oxidation. A schematic inset isincluded to indicate the relative depth of decay of the interstitial profile (~ tens ofmicrons) versus the diffusion length of the boron profiles examined in the near surface(top 1 micron).

Because of the technological demand for smaller device structures, it is

increasingly important to be able to reduce the boron diffusion length for any given

transistor fabrication process. This has traditionally been done by reducing the time

and temperature of all thermal treatments in the fabrication process, i.e. using a

smaller “thermal budget”. However, efforts to shrink the total boron diffusion length

in the presence of an oxidation step are frustrated by the exponential increase in boron

OED as the temperature is lowered, Fig. 3.2 (a). Clearly, a much lower temperature is

required to achieve the same total diffusion lengths, when OED effects are present,

1017

1018

1019

1020

1021

0 0.2 0.4 0.6 0.8 1

Con

cent

rati

on[c

m-3

]

Depth [µm]

BB B B

As-Grown

O2

N2

750°C240 min

0

5

10

15

20

25

30

35

0 200 400 600 800 1000Boron Marker Depth [nm]

OE

DE

nhan

cem

ent [

D/D

* ]

750°C240 min

(a) (b)

LDB

∆D ~ 0

Depth

[ I* ]

[ IOED ]

[I

]

54

because of the smaller effective activation energy of the oxidation enhanced boron dif-

fusivity, Fig. 3.2 (b). The primary point is that the boron diffusivity is sensitive to

excess interstitials introduced by fabrication steps and that at lower temperatures the

extra interstitials become the dominant source of boron diffusion. To further compli-

cate matters, the absolute interstitial concentrations and diffusivity are notoriously dif-

ficult to directly measure and remain experimental unknowns [11]. Therefore, despite

extensive research on silicon oxidation, a detailed quantitative understanding of the

microscopic interstitial defect injection and formation physics does not exist [20].

3.2.3 Transient enhanced diffusion

As was discussed in the previous section, boron and phosphorus diffusion is enhanced

during oxidation due to excess interstitials produced from the chemical reaction at the

surface. Another important example of enhanced boron diffusion due to processing is

transient enhanced diffusion (TED), which is the enhanced diffusion observed imme-

diately after implantation of dopants. As is easy to imagine, the implantation process

is very destructive to the silicon crystal lattice and therefore produces numerous point

defects. After implantation it is necessary to anneal to recrystallize any silicon driven

amorphous and electrically activate the implanted dopants. Boron diffusivities imme-

diately after implantation have been reported to be thousands of times greater than the

intrinsic diffusivities [21], due to the excess interstitial population created by the

implantation. Furthermore, unimplanted dopants near implanted regions also experi-

55

ence similarly enhanced diffusivities, because the point defects created by the implant

rapidly migrate to all nearby regions [21].

Because the enhanced diffusion from extra interstitials caused by processing

dominate diffusion at lower temperatures, common steps like ion implantation (TED)

and oxidation (OED) are recognized as significant challenges to the increased perfor-

mance (scaling) of sub-micron devices [22]. In the next section the importance of con-

trolling boron diffusion and reducing TED effects will be considered in the specific

example of the SiGe HBT.

3.3 Boron diffusion in SiGeC

3.3.1 Introduction of Cs to reduce boron diffusion

In n-Si/p+SiGeC/n-Si HBTs, as boron diffuses from the p+SiGeC base into n-Si

emitter and collector (Fig. 3.5), parasitic barriers are formed in the conduction band,

which impede the flow of electrons from the emitter to collector [23, 24] (Fig. 3.6

(a)). The parasitic barrier that arises due to boron outdiffusion is strongly dependent

on the boron concentration that diffuses into the silicon, and small amounts of boron

outdiffusion Ld ~ 10 Å can already cause large parasitic barriers evident in HBT's [24]

because the collector current is exponentially dependent on the barrier height, sche-

matically shown in Fig. 3.6 (b). This can be observed by directly measuring collector

saturation current, or even more sensitively by observing the effect of collector-emitter

bias on collector current (the Early effect). The Early voltage changes because the

56

barrier height, and hence the collector current, is affected by the collector-emitter bias

[25].

Figure 3.5. Schematic diagram of (a) the as-grown dopant profiles of a SiGe HBTstructure and (b) the profile after annealing. Boron is shown to diffuse faster than thesurrounding germanium (SiGe) layer forming a p-type region in the phosphorus dopedsilicon region.

Recently, the suppression of boron diffusion of both thermal and transient

enhanced diffusion (TED) has been demonstrated through the incorporation of substi-

tutional carbon in pure silicon and SiGe increasing the thermal budgets of the fabrica-

tion processes significantly [23, 26-28]. Because HBTs are so sensitive to boron

diffusion from the base on a scale less than that detectable by SIMS, the HBT is an

ideal probe to examine the effect that substitutional carbon has on boron diffusion.

Therefore, we use HBT electrical characteristics to quantitatively compare boron dif-

fusion in SiGe to boron diffusion in SiGeC for annealing or implant and annealing

p+n+

n-

As-grown

p

ActivationAnneal

Con

cent

rati

on

Depth

Boron

SiGe layersurrounds boron

pp+

SiGe layer

Annealed Boron outsideSiGe layer

n-n+

Depth

Con

cent

rati

on

(a) (b)

57

conditions. Because the devices are greatly affected by diffusion at levels too small to

be detected by SIMS, we use modeling of combined process and device simulators

(TMA TSUPREM4 and MEDICI) to relate observed electrical characteristics (collec-

tor saturation currents and Early voltages) of the HBT's to infer the changes in boron

profile and hence the changes in boron diffusion coefficients.

Figure 3.6 (a) schematic conduction band diagram of a (n-)Si/(p+)SiGeC/(n-)Si HBTas grown and after annealing showing the creation of a parasitic conduction band bar-rier as a result of boron diffusion from the base into the n-type Si emitter and collectorregion, and (b) the corresponding schematic reduction of collector current afterannealing [24].

3.3.2 HBT Process Conditions

The HBT's were grown by RTCVD [29] at 575-700°C, with boron levels in the base of

~1020 cm-3 and bases of ~20 nm of Si0.8Ge0.2 or Si0.795Ge0.2C0.005. The device fabri-

cation was done using all low-temperature processing to avoid unnecessary diffusion

and is described elsewhere [23, 26]. Photoluminescence and X-ray diffraction studies

0

0.2

0.4

0.6

0.8

1

0.15 0.2 0.25 0.3 0.35

Con

duct

ion

Ban

dE

nerg

y[e

V]

Depth [µm]

10-9

10-8

10-7

10-6

10-5

0.0001

0 0.1 0.2 0.3

Col

lect

orcu

rren

tden

sity

[Acm

-2]

Base-emitter voltage [V]

As-Grown

Annealed

As-

Gro

wn

Ann

eale

d

p+ basen- n-

(a) (b)

SiGeSi Si

Parasitic Barriers

58

on similar alloy layers show that the alloy layers are biaxially compressively strained

to match the silicon lattice, and transmission electron microscopy (TEM) showed no

dislocations, defects, or SiC precipitates in any of the as-grown layers [23, 26]. As-

grown Gummel plots and common emitter characteristics of HBT's without and with

0.5% substitutional carbon in the bases are shown in Fig. 3.7 (a) and (b) respectively.

High Early voltages and SIMS verifies that there is no significant boron outdiffusion

[23, 26] in such HBT's fabricated without annealing.

Figure 3.7 HBT Gummel plots and collector current vs. base-collector voltage afterimplant into 2000 Å Si n- emitter and 647°C anneal for HBT’s with (a) SiGe or (b)SiGeC bases [23].

Decreasein Ic

Low VA

High VA

Littledecrease

in Ic

(a) (b)

59

Two different cases are considered for HBT processing to study boron outdif-

fusion. Case (1): the effect that substitutional carbon has on the intrinsic boron diffu-

sion rates (N2 anneal, 15 minutes, 800-950°C); and case (2): The effect of

substitutional carbon on the transient enhanced diffusion of boron due to ion implant

damage in the overlying emitter layer (1.5x1015 cm-2, As+, 30 keV and 3x1014 cm-2,

As+, 15 keV into the silicon 2000 Å n- emitter) with subsequent 15 minute activation

anneals in N2 at 647°C and 742°C. Note the range of the arsenic implant was at most

50 nm, so all damage occurs well above the base region.

SIMS, Gummel plots and common emitter characteristics of the processed

HBT's with ion implant damage are shown in (Fig. 3.7). Saturation currents and Early

voltages were then extracted from the electrical characteristics for comparison and fit-

ting to simulated electrical characteristics. The decrease in Ic and reduced Early volt-

ages in the transistors annealed at 647°C without carbon, Fig. 3.7 (a), show that boron

has outdiffused significantly during this step even though this annealing condition is

far less than the emitter thermal budget (i.e. the thermal budget required to grow the

emitter by RTCVD). However the high Early voltages in the HBT devices with car-

bon show that much of the TED effects have been suppressed. Even in this case the

Early voltage is not as high as that of the as grown HBT, ~10.5 V, evidence that some

slight TED effects still remain, despite no evidence of boron outdiffusion at 647°C in

boron profiles obtained using SIMS. Figure 3.8 shows that boron outdiffusion is

60

readily apparent for As implantation and a 755°C, 15 min N2 anneal in the transistor

without carbon, but is substantially reduced in the transistor with carbon.

Figure 3.8 Boron, carbon and germanium profiles of the (a) SiGe and (b) SiGeCHBT’s after implant into emitter and anneal at 755°C in N2 for 15 minute [23, 26].Note: no difference in boron concentration profiles of the implanted and 647°Cannealed HBTs was observed from the SIMS profiles.

3.3.3 Extraction of diffusion constant (method)

Dopant profiles were modeled using TSUPREM4 (TMA), and the results were

used in the device simulator MEDICI (TMA). The diffusion coefficient of boron in

SiGeC (case 1) and excess interstitial concentration (case 2) were adjusted in

TSUPREM4 so that the output of the device simulator matched the experimentally

T=755ºC15 min

T=755ºC15 min

Base

Base

Ge

Ge

B

C

B

(a)

(b)

61

measured Gummel plots and common-emitter characteristics. In case 1 a single effec-

tive diffusion constant, Deff, for boron was used to fit electrical data for the entire

device structure, maintaining the ratio of neutral and singly charged defect contribu-

tions to the diffusion constant and only varying the default magnitude by a constant.

In case 2 (arsenic implant), the number of excess interstitials resulting from implant

damage leading to enhanced boron diffusion was scaled to match the experimentally

extracted electrical characteristics, while using a typical diffusion constant of boron in

silicon at 647°C of 9.33x10-10 um2 min-1 corresponding to a boron diffusion length of

approximately 1 Å after 15 minutes.

The HBT electrical characteristics were numerically simulated using the dop-

ing profiles obtained above to make comparison to experimental data. Bandgap differ-

ences of 160 meV and 147 meV were used for SiGe and SiGeC bases respectively.

The effective density of states in the Si1-xGex base, approximately

for x=0.2, is assumed not to change in the SiGeC base; and a bandgap narrowing

model commensurate with observed bandgap narrowing in SiGe due to high doping

densities in SiGe [30], which is less than that observed in Si, was also included.

3.3.4 Boron diffusivity found in SiGe and SiGeC HBT bases

Collector saturation currents (y axis intercept) extracted from gummel plots (collector

current vs. base emitter voltage) of fabricated HBT's for case 1, intrinsic diffusion, are

33.0≅SiVC

SiGeVC

NN

NN

62

shown in Fig. 3.9. Typical boron diffusion lengths in silicon at 855°C for 15 minutes

are ~75 Å [2]. The saturation current of the Si/SiGe HBT's annealed at 855°C is

already reduced nearly two orders of magnitude demonstrating the extreme HBT sen-

sitivity to small boron diffusion lengths. For HBT processing this sensitivity is unde-

sirable because it limits the total thermal budget available to the process engineer.

Through the addition of substitutional carbon the onset of saturation current degrada-

tion can be shifted to higher temperatures increasing the available thermal budget, in

this case, by as much as ~100°C.

Figure 3.9 HBT saturation currents extracted from Gummel plots of fabricated andnumerically simulated devices. Fabricated devices are indicated by solid markers,numerically simulated by hollow markers. Note carbon incorporation increases ther-mal budget of HBT process ~100°C.

10-11

10-10

10-9

650 700 750 800 850 900 950 1000

SiGe (simulated)SiGeC (simulated)SiGe (lab)SiGeC (lab)

Col

lect

orSa

tura

tion

Cur

rent

[Acm

-2]

Temperature [Celsius]

SiGe

SiGeC

63

To quantitatively estimate the relative boron diffusion constants in SiGe and

SiGeC, the experimentally observed collector saturation currents were numerically

calculated and fit to the observed data by varying a single diffusion parameter, the rel-

ative interstitial concentration, as described above (eq’n. 3.11). The numerically

obtained diffusion constants for boron in the SiGe and SiGeC are compared with the

activation energy of boron diffusion in silicon [2] in Fig. 3.10. The fitted boron diffu-

sivities in SiGeC are uniformly slower than those in Si and SiGe. The best fits yield

boron diffusivities in SiGeC that are ~8 times less than that in SiGe and appear to have

a similar activation energy as that of pure silicon, showing that the incorporation of

carbon does not significantly change the energetics of boron diffusion compared to sil-

icon. This also agrees with previously reported boron diffusivities in SiGe and SiGeC

that find boron to move slower in SiGeC than in SiGe and both diffusivities to be

slower than that in silicon [31-33], but the absolute diffusivities extracted from the

HBT data are approximately 2-4 times faster than those reported. Various sources of

error can contribute to disagreement with the referenced values. The numerically cal-

culated profiles are simulated with a single diffusion constant for boron. The extracted

diffusivity will represent an average diffusivity of that in the alloy layer and that in sil-

icon. For long diffusion lengths, with respect to the width of the HBT base, the

numerically found boron diffusivity in the alloy should be faster than that of the actual

64

boron diffusivities in the alloy. Other sources of the disagreement between the

reported values and ours can come from errors in temperature calibration or not com-

pensating for interface diffusion effects like the built in electric field. The combined

effects can easily lead to factors of two in diffusivity.

Figure 3.10 Boron diffusion constants obtained from fitted collector saturation cur-rents in figure 3.8. Note boron diffusivity in SiGeC ~8 times less than that in SiGe.Reference line shows activation energy of 3.4 eV.

To illustrate the HBT sensitivity to boron diffusion, the extracted diffusion

constant for boron in the 855°C annealed SiGeC device can be used to calculate an

approximate boron diffusion length of ~25 Å [1]. This diffusion length is discernibly

signaled by a one-half drop in collector saturation current compared to that of the as-

10-17

10-16

10-15

0.8 0.85 0.9 0.95

Dif

fusi

onC

onst

ant[

cm2

s-1]

1000/T [Kelvin-1]

~ 3.4 eV

800ºC

850ºC

950ºC SiGe

SiGeC

65

grown HBT. However, such a small diffusion length would be nearly impossible to

resolve by SIMS since broadening of the boron profile in SIMS can be of the order of

20-40 Å [31]. Thus HBT electrical characteristics are more sensitive than SIMS to

small diffusion lengths of boron. This highlights the necessity of minimal boron diffu-

sion to maintain optimum performance of the SiGe HBT.

For case 2 (ion implant damage in overlying emitter layer), Early voltages

were extracted from both fabricated devices, Fig. 3.11, and calculated using numerical

simulations of the arsenic implant and anneal at 647°C. Excess interstitials due to ion

implant damage lead to TED of boron and degrade device performance. The total

excess interstitial concentrations were adjusted as the single parameter to fit the

observed Early voltages of the devices. The unadjusted model estimates the excess

interstitial concentration from implant damage to be ~1.31x1014 cm-2 for an implant of

30 keV, 1.5x1015 cm-2 As+ followed by 15 keV, 3x1014 cm-2 As+. The resulting diffu-

sion causes a predicted Early voltage of 0.3 V agreeing well with experiment for the

SiGe case. In the case of a SiGeC base, the calculated Early voltage could only be

made to agree with the observed device Early voltage after a reduction of over 99% of

the excess interstitials [1]. This corresponds to a diffusion length of only a few Ång-

stroms, again showing the sensitivity of the device to minute amounts of boron diffu-

sion, while no difference in SIMS profiles in that case could be seen.

66

Figure 3.11 Calculated Early voltages for adjusted excess interstitials due to implantdamage. Experimentally observed Early voltages obtained from HBT common-emit-ter electrical characteristics for asgrown HBTs with SiGeC bases, and implant andannealed HBTs with SiGe and SiGeC bases are indicated by circles.

It can be concluded that the SiGe(C) HBT is extremely sensitive to boron dif-

fusion lengths as small as 10 Å. Furthermore, the introduction of substitutional carbon

clearly reduces boron diffusion and thereby increases the thermal budget for the fabri-

cation of SiGe HBT. The intrinsic boron diffusivity in Si0.795Ge0.2C0.005 is estimated

to be 8 times less in Si0.8Ge0.2 and the substitutional carbon has been shown to sink

approximately 99% of the interstitials that reach the SiGeC layer from the ion implant

damage.

1

10

10-8 10-6 0.0001 0.01 1

Ear

lyV

olta

ge[|V

|]

Relative Excess Interstitial Concentration

IntrinsicDiffusion only

SiGeCbase

SiGebase

67

3.4 Summary

The microscopic mechanisms of solid-state boron diffusion in silicon were examined

and boron diffusivity was shown to be linearly dependent on the silicon interstitial

concentration. Because current transistor dimensions require extreme control over the

boron profiles, common fabrication steps that produce extra silicon interstitials in the

silicon lattice, e.g. ion implantation and oxidation, represent significant challenges to

fabricating high performance transistors. Specifically, even in the case of ultra small

thermal budgets (647°C, 15 minutes, LD ~ 1 Å) boron TED still leads to noticeable

decreases in the SiGe HBT’s Early voltage and saturation currents, demonstrating the

extreme sensitivity of the transistor to excess interstitials. The introduction of 0.5%

substitutional carbon to the SiGe base is shown to reduce the effective intrinsic boron

diffusivity by as much as a factor of 8 and sinks almost all excess interstitials that

reach the SiGeC layer for this process. The success of substitutional carbon in reduc-

ing the boron diffusivity when introduced to the base of SiGe motivates the following

chapters on determining what mechanism is responsible for the reduced boron diffu-

sion when carbon is present. Some outstanding questions remaining to be answered

are:

(1) Why does the carbon reduce the boron diffusivity?

(2) Is the lower diffusivity observed in SiGe vs. that in Si due to the same

mechanism?

68

(3) Does the carbon react with the boron directly?

(4) Does the carbon react with the interstitials directly?

(5) How much carbon is needed, i.e. what is the carbon concentration

dependence?

(6) Do carbon related defects form?

These critical questions are addressed in the following chapters.

69

3.5 References

[1] M. S. Carroll, L. D. Lanzerotti, and J. C. Sturm, “Quantitative measurement ofreduction of boron diffusion by substitutional carbon incorporation,” presented at Dif-fusion Mechanisms in Crystalline Materials, San Francisco, 1998.[2] Fahey, P. B. Griffin, and J. D. Plummer, Rev. Mod. Phys. , vol. 61, pp. 289,1989.[3] M. R. Pinto, D. M. Boulin, C. S. Rafferty, R. K. Smith, W. M. Coughran, I. C.Kizilyalli, and M. J. Thoma, Tech. Digest IEDM , pp. 923, 1992.[4] G. D. Watkins, Phys. Rev. B , vol. 12, pp. 5824, 1975.[5] R. D. Harris, J. L. Newton, and G. D. Watkins, “Negative-U defect: Interstitialboron in silicon,” Phys. Rev. B , vol. 36, pp. 1094, 1987.[6] H.-J. Gossmann, T. E. Haynes, P. A. Stolk, D. C. Jacobson, G. H. Gilmer, J. M.Poate, H. S. Luftman, T. K. Mogi, and M. O. Thompson, Appl. Phys. Lett. , vol. 71, pp.3862, 1997.[7] A. Ural, P. B. Griffin, and J. D. Plummer, “Fractional contributions of micro-scopic diffusion mechanisms for common dopants and self-diffusion in silicon,” J.Appl. Phys. , vol. 85, pp. 6440, 1999.[8] R. B. Fair and P. N. Pappas, J. Electrochem. Soc. , vol. 122, pp. 1241, 1975.[9] R. F. Lever, J. M. Bonar, and A. F. W. Willoughby, J. Appl. Phys. , vol. 83, pp.1988, 1998.[10] T. Y. Tan and U. Goesele, Appl. Phys. A , vol. 37, pp. 1, 1985.[11] H. Bracht, MRS Bulletin , vol. 25, pp. 22, 2000.[12] W. Windl, M. M. Bunea, R. Stumpf, S. T. Dunham, and M. P. Masquelier,Phys. Rev. Lett. , vol. 83, pp. 4345, 1999.[13] M. E. Law, C. S. Rafferty, and R. W. Dutton, “SUPREM IV User's Manual,” .Stanford, CA: Stanford University, 1988.[14] Muraka, J. Appl. Phys. , vol. 48, pp. 5020, 1977.[15] D. Tsoukalas, D. Skarlatos, and J. Stoemenos, J. Appl. Phys. , vol. 87, pp.8380, 2000.[16] S. T. Dunham and J. D. Plummer, “Point-Defect Generation during Oxidationof Silicon in Dry Oxygen Part 1: Theory,” J. Appl. Phys. , vol. 59, pp. 2541, 1986.[17] G. Charitat and A. Martinez, J. Appl. Phys. , vol. 55, pp. 909, 1984.[18] W. Tiller, J. Electrochem. Soc. , vol. 127, pp. 625, 1980.[19] Y. Shibata, K. Taniguchi, and C. Hamaguchi, J. Appl. Phys. , vol. 70, 1991.[20] S. T. Dunham and J. D. Plummer, “Interactions of silicon point defects withSiO2 films,” J. Appl. Phys. , vol. 71, pp. 685, 1992.

[21] P. A. Stolk, H.-J. Gossmann, D. J. Eaglesham, D. C. Jacobson, C. S. Rafferty,G. H. Gilmer, M. Jaraiz, J. M. Poate, H. S. Luftmann, and T. E. Haynes, “Physical

70

mechanisms of transient enhanced dopant diffusion in ion-implanted silicon,” J. Appl.Phys. , vol. 81, pp. 6031, 1997.[22] P. Packan, MRS Bulletin , vol. 25, pp. 18, 2000.[23] L. D. Lanzerotti, J. C. Sturm, E. Stach, R. Hull, T. Buyuklimanli, and C.Magee, Appl. Phys. Lett. , vol. 70, pp. 23, 1997.[24] E. J. Prinz, P. M. Garone, P. V. Schwartz, X. Xiao, and J. C. Sturm, IEEE Elec.Dev. Lett. , vol. 12, pp. 42, 1991.[25] E. J. Prinz and J. C. Sturm, “Current Gain-Early Voltage Products in Hetero-junction Bipolar Transistors with Nonuniform Base Bandgaps,” IEEE Elect. Dev. Lett., vol. 12, pp. 661, 1991.[26] L. D. Lanzerotti, J. C. Sturm, E. Stach, R. Hull, T. Buyuklimanli, and C.Magee, IEDM Tech. Digest , pp. 249, 1996.[27] R. Scholz, U. Goesele, J. Y. Huh, and T. Y. Tan, Appl. Phys. Lett. , vol. 72, pp.2, 1998.[28] H. J. Osten, G. Lippert, P. Gaworzewski, and R. Sorge, Appl. Phys. Lett. , vol.71, pp. 11, 1997.[29] J. C. Sturm, H. Manoharan, L. Lenchyshyn, M. Thewalt, N. Rowell, J. P. Noel,and D. Houghton, Phys. Rev. Lett. , vol. 66, pp. 1362, 1991.[30] Z. Matutinovic-Krstelj, V. Venkataraman, E. J. Prinz, and J. C. Sturm, IEEETrans. Elec. Dev. , vol. 43, 1996.[31] P. Kuo, J. L. Hoyt, Gibbons, Turner, Jacowitz, and T. Kamins, Appl. Phys. Lett., vol. 62, 1993.[32] A. F. W. Willoughby, J. M. Bonar, and A. D. N. Paine, “Diffusion in SiGeAlloys,” presented at Si Front-End Processing-Physics and Technology of Dopant-Defect Interactions, San Fransisco, 1999.[33] N. Larsen, “Impurity Diffusion in SiGe Alloys: Strain and CompositionEffects,” presented at Diffusion Mechanisms in Crystalline Materials, San Fransisco,1998.[34] Roth and Plummer, J. Electrochem. Soc. , vol. 141, pp. 1074, 1994.[35] S. Pindl, M. Biebl, and E. Hammerl, J. Electrochemical Soc. , vol. 144, pp.4022, 1997.[36] P. Packan and J. D. Plummer, J. Appl. Phys. , vol. 168, pp. 4327, 1990.

71

Chapter 4

Substitutional Carbon as a Silicon Self-Interstitial Sink

4.1 Introduction

In the previous chapter the importance of limiting boron diffusion during processing

was highlighted by the SiGe HBT’s electrical characteristics, which are sensitive to

boron diffusion lengths as small as 1-2 nm. One novel approach to reduce boron diffu-

sion, discussed in the previous chapter, is the introduction of substitutional carbon into

the SiGe base region. The incorporation of 0.5% carbon was found to be enough to

reduce the intrinsic base boron diffusivity by as much as a factor of 8 and to eliminate

~99% of the boron diffusivity enhancement from the excess interstitials created by ion

implant damage.

Because of the possible technological advantages of substitutional carbon incor-

poration a deeper understanding of the point defect relationship with the substitutional

carbon is desirable in order to predict critical variables like the necessary carbon con-

centrations required to eliminate OED and TED effects in a device structure. In this

chapter, a direct reaction between substitutional carbon and the silicon self-interstitials

is established. In the following chapters the carbon-self-interstitial reaction in SiGe is

quantified.

72

4.2 Previous reports on reduced boron diffusivity in the presence of carbon

As discussed previously, it is generally accepted that boron diffuses primarily via an

interstitial mechanism and that the boron diffusivity is linearly dependent on the sili-

con interstitial concentration. Ion implant damage and oxidation are sources of excess

interstitials, which enhance the boron diffusivity, therefore the reduction of TED and

OED depend on the effective ability to trap or annihilate any excess interstitials that

are produced during processing. The boron diffusivity has already been observed to

strongly depend on the concentration of substitutional carbon in silicon [1-4]. Various

approaches of carbon incorporation into silicon have been used to explore the carbon

effect on enhanced boron diffusion including carbon-containing alloys (e.g. SixCy [5]

and Si1-x-yGexCy [1, 6]), uniform carbon doping of silicon [2], and localized carbon

doping of silicon [7]. These approaches may be summarized in to two categories.

First, silicon or carbon-silicon alloys that have a uniform distribution of carbon in

them [5, 7]; and second, silicon with localized carbon doped regions or thin carbon

containing alloy layers [6, 7]. Complete suppression of the TED and OED of boron

has already been reported for the category of uniform carbon concentration [7]. How-

ever, for practical use in device applications this approach is hampered by concerns

about the electrical activity of carbon-related defects in silicon [6, 8, 9].

Localized carbon doped regions, however, offer a potential solution to the dif-

ficulties presented by uniform carbon doping. Carbon containing regions may be

placed away from the active region eliminating concerns about electrical defects, and

73

also creating a test structure with which to isolate carbon’s effect on interstitial con-

centration without the complication of carbon or germanium’s effects on the intrinsic

boron diffusivity. However, no complete suppression of TED or OED was reported

before this work with a localized carbon region. This work was reported in the follow-

ing publications [10-12].

4.3 Interstitial filter experiment using SiGeC

The test structures were grown using rapid thermal chemical vapor deposition

(RTCVD) [13], between 600 and 750°C using methylsilane as the carbon source. A

phosphorus layer at the surface and two boron doped silicon layers with or without a

Si1-x-yGexCy or Si1-xGex layer placed between the boron-doped layers were grown

above a boron doped buffer layer ~ 1 µm thick. The structure is used to test the effect

of the Si1-x-yGexCy layer on phosphorus and boron diffusivities at different locations

(above and below) with respect to the SiGe(C) layer. The phosphorus layer concentra-

tion was ~1x1019 cm-3 and approximately 1500 Å thick, and both boron peaks were

approximately 250 Å thick and had a boron concentration of ~5x1019 cm-3 centered

2000 Å and 3000 Å away from the surface respectively, while the 250 Å thick SiGe(C)

layer was centered 2300 Å from the surface (Fig. 4.1).

74

Figure 4.1 Test structure to study boron or phosphorus oxidation enhanced diffusionand transient enhanced diffusion. Two boron markers are separated by a Si1-x-yGexCyinterstitial barrier layer (x = [0 - 0.2]; y = [0 - 0.005])

To study boron OED dependence on substitutional carbon level

as-grown samples were cleaved and annealed in nitrogen and oxygen ambient atmo-

spheres for 30 minutes at 897°C. For this experiment, the temperature measured by

the thermocouple in the annealing furnace are suspect because of poor alignment. The

thermocouple was located improperly close to the cold-zone and therefore indicated

an inaccurately low temperature. The temperature of the sample was determined,

alternatively, by calculating the boron diffusivity from the annealing time and the mea-

Intersititials & oxide forms during oxidationI

i-Si

Si : boron 3x1019/cm-3

i-Si

i-Si1-x-yGexCy

Si buffer layer: boron 3x1019/cm-3

i-Si

i-Si

I250 Å

250 Å

250 Å

2000 Å

100 Å

400 Å

600 ÅSi : phosphorus 1x1019/cm-3

Si : boron 3x1019/cm-3

1000 Å

75

sured boron diffusion length (using SIMS) in the pure silicon sample. The tempera-

ture of the sample was then deduced by comparing the well established temperature

dependence of the boron diffusivity in silicon [14] to the measured boron diffusivity in

the silicon sample. The deduced temperature also agrees with the resulting oxide

thickness and amount of oxidation enhanced diffusion observed for these samples.

Boron profiles were characterized using secondary ion mass spectroscopy (SIMS)

with 2 keV Cs+ ions a 15-20% error in boron and phosphorus concentrations and an

estimated 1-5% error in depth scales. Figure 4.2 shows profiles of the boron concen-

tration in a pure silicon sample, i.e. no SiGe(C) layer, with a background carbon con-

centration below SIMS detection limits. Broadening of both boron peaks is observed

after the nitrogen anneal, and in the oxidation case clear enhancement of the boron dif-

fusion is observed by the further broadening of the boron profile. Further evidence of

the OED is the significant boron diffusion from the buffer layer that has diffused over

100 nm and now interferes with the boron profile of the bottom most marker layer.

76

Figure 4.2 Boron concentration profiles (measured using SIMS) from samples ofdoped pure silicon (no intervening SiGeC layer) of as-grown and annealed for 897°Cfor 30 minutes in nitrogen ambient, or in oxygen ambient samples. Oxidationenhanced diffusion (OED) is demonstrated by the broader boron profile after anneal inoxygen ambient compared to that annealed in nitrogen.

To quantitatively compare the effect of different annealing conditions on the

boron, boron diffusivities for each profile were obtained by fitting each concentration

profile with the process simulator PROPHET [15] using a single variable fit (i.e. the

ratio of excess self-interstitial to intrinsic self-interstitial concentrations I / I*). In this

case (Fig. 4.2), the boron diffusivity during oxidation is found to be 6 times that of the

average boron diffusivity during the nitrogen anneal. Figure 4.3 shows boron profiles

before and after anneals for the sample containing an intervening thin 250 Å

1017

1018

1019

0.1 0.2 0.3 0.4

Bor

onC

once

ntra

tion

[cm

-3]

Depth [µm]

T=897ºC30 min

As-Grown

N2

O2

From buffer layer

77

Si0.795Ge0.2C0.005 layer between the boron markers. Two dashed lines are over-laid to

indicate the location of the SiGeC layer. The same broadening is observed in the peak

below the Si1-x-yGexCy layer after annealing in nitrogen, but in this case there is no

additional broadening after oxidation, indicating that for this substitutional carbon

level the oxidation-enhanced diffusion below the SiGeC has been completely sup-

pressed. The excess interstitial concentration, therefore, has been reduced to zero

below the carbon containing layer and shows that the Si1-x-yGexCy layer blocks inter-

stitials from reaching the layer below completely suppressing boron OED.

We also conclude from this sample that the formation of an immobile complex

like Bi-Cs is not responsible for the “sinking” of the injected self-interstitials.

Although, in this case, some boron do migrate into the SiGeC layer and hypothetically

might form immobile complexes trapping some self-interstitials this can not explain

the “sinking” effect. Most interstitials pass through the top most boron layer, indicated

by the similar enhanced diffusion in both boron peaks in the pure-silicon sample.

Since the Bi that form and migrate to the carbon layer in the carbon-containing sample

represent only a small fraction of all the self-interstitials passing through the top boron

marker layer, the subsequent hypothetical immobile Bi-Cs cannot be responsible for

trapping an appreciable number of the injected interstitials. Therefore, a Bi-Cs com-

plex can be ruled out as the primary interstitial trap.

78

Figure 4.3 Boron concentration profiles from samples of doped silicon with the high-est level of carbon in the SiGeC barrier layer of as-grown samples, and samplesannealed for 897°C for 30 minutes in nitrogen ambient, or in oxygen ambient sampleswith. Boron OED is completely suppressed as indicated by no difference betweenboron profiles of the deepest marker after anneal in oxygen and anneal in nitrogenambient.

The boron profiles of the boron marker above the Si0.795Ge0.2C0.005 are also

nearly identical after anneals in either oxygen or nitrogen, Fig. 4.3, indicating that the

excess interstitial concentration must also be reduced to nearly zero even as far as 100-

200 Å above the intervening Si1-x-yGexCy layer. The boron profile above the SiGeC

layer after annealing is unusual because of the asymmetric broadening, spiked profile

at the Si/SiGeC interface, and the overall peak shift towards the SiGeC layer. The

10 17

10 18

10 19

10 20

0.1 0.2 0.3 0.4 0.5

Bor

onC

once

ntra

tion

[cm

-3]

Depth [µm]

T=897ºC30 min

As-Grown

N2

O2

As-Grown

SiGeC

79

boron profile in the upper marker layer is not understood at this time but the behavior

might be attributed to the combined effects of diffusion in an interstitial concentration

gradient near the Si1-x-yGexCy layer (see next section and following chapter), diffusion

into a Si1-x-yGexCy region of much slower boron diffusivity (previous chapter), and

boron segregation into the Si1-x-yGexCy alloy layer (independent of the diffusivity dif-

ferences). Boron segregation into strained Si0.8Ge0.2 from silicon in a ratio of 1.2 at

923ºC with an activation energy of 0.3 eV has previously been reported, [16, 17] and

further study is in progress to determine the boron segregation and diffusivity depen-

dence on carbon content.

Samples with different substitutional carbon levels in the Si1-xGex alloy layer

were grown and annealed to test the OED carbon level dependence. Figure 4.4 sum-

marizes the relative boron diffusivities of the boron peak under the Si1-x-yGexCy dur-

ing anneals in oxygen versus nitrogen ambient, for carbon levels of 0 to 0.5%. To

within experimental error there is no difference between the OED of boron in the cases

of boron markers in pure silicon and that with an intervening Si1-xGex alloy layer (no

carbon) between the boron markers. This demonstrates that the interstitial trapping is

not a result of the addition of the Si1-xGex alloy layer and is in agreement with previ-

ous work [6, 18]. Finally, we observe a monotonic decrease in boron OED with

increasing carbon level in the alloy layer indicating that as the carbon level is

increased more interstitials from the surface region react with the carbon and fewer

80

interstitials pass through the layer, schematically represented in Fig. 4.5. The total car-

bon level required to completely suppress the OED for these conditions is bound

between 4.56x1013 cm-2 and 5.57x1014 cm-2 integrated carbon level.

Figure 4.4 Average diffusivity (boron) enhancement’s dependence on carbon levels inSiGe(C) barrier layer for oxidation 897°C for 30 min. Boron diffusivity of deepestmarker is displayed relative to that of the average boron diffusivity during anneal innitrogen ambient, 897°C for 30 min in nitrogen. The schematic diagram below the y-axis label is to indicate that these are the diffusivities of the bottom most boron markerlayer.

The dependence of boron TED on substitutional carbon level

as-grown samples (Fig. 4.1) with and without buried Si1-x-yGexCy layers were cleaved

and subjected to implant levels of 5x1013 cm-2 and 5x1014 cm-2 silicon with an

2.5

5

7.5

Si SiGe Carbon4.56x1013/cm2

Carbon5.57x1014/cm2

T=897ºC30 min

No OED1

Bor

onD

iffu

sivi

tyE

nhan

cem

ent

(DO

x/D

N2)

SiGeC

B B

81

implant energies of 30 keV, and then annealed in a nitrogen ambient for 15 minutes at

800°C. The depth of the silicon implanted profile is estimated to be 600 Å. Migration

of interstitials from the ion implant damaged region lead to excess diffusion of the

lower boron peak (as measured by SIMS), which decreases as the carbon level of the

barrier layer is increased. At the highest carbon levels no significant difference in the

lower boron profiles can be seen between samples with or without ion implant dam-

age, again indicating complete filtering of the excess interstitials. Average boron dif-

fusion constants were then extracted by fitting measured boron profiles to simulation.

Figure 4.6 summarizes, for two different silicon implant doses and different carbon

levels in the barrier level, the average boron diffusivity after ion implantation relative

to the measured boron diffusivity without silicon ion implantation in the same sam-

ples. For both implant doses of 5x1013 cm-2 and 5x1014 cm-2 silicon the total carbon

level required to completely suppress the TED for these conditions is bound between

4.56x1013 cm-2 and 5.57x1014 cm-2 carbon level.

82

Figure 4.5 Schematic diagram of the interstitial concentration for three cases of car-bon content in the buried SiGeC layer. As carbon level is increased, the interstitialconcentration in both the surface region and below the SiGeC layer are reduced. Theexact profile is yet to be determined.

For the case when the surface is not driven amorphous, implantation has been

found to create a little over one excess interstitial for every implanted ion, which are

known to condense into interstitial clusters for similar conditions [7]. An accurate

estimate of the number of interstitials that are released and migrate to the carbon layer

is not known at this time because detailed knowledge about the dynamics of how the

interstitials are released and how many are annihilated at the nearby surface is not

known. Note: the boron TED after implantation with the higher dose is only a little

over three times as great as the lower dose although 10 times more silicon is

implanted. In this case, the surface is driven amorphous and the number of excess

BExc

ess

[I

]

0.5%

0.05%

0.0%

Depth

SiGeC

83

interstitials created by the implant no longer grows linearly with dose because the

interstitials formed in the amorphous layer are annihilated during the subsequent solid-

phase epitaxy.

Figure 4.6 Average boron diffusivity (during 800°C 15 minute anneal after ion

implantation for Si doses of 5x1013/cm2 or 5x1014/cm2 at 30 keV) dependence on car-bon levels in SiGe(C) barrier layer. Boron diffusivities of deepest marker is displayedrelative to that of the average boron diffusivity during an 800°C 15 minute anneal innitrogen ambient atmosphere without ion implantation. The schematic diagram belowthe y-axis label is to indicate that these are the diffusivities of the bottom most boronmarker layer.

4.3 Phosphorus diffusion above the SiGeC layer

Figure 4.7 shows profiles of the overlying phosphorus edge in the all-silicon control

sample, with a background carbon concentration below SIMS detection limits. Broad-

ening of the phosphorus edge is observed after the nitrogen anneal, and in the oxida-

1

10

100

10 111012 10

1310

1410

15

Integrated carbon level [cm-2]

0

5x1013/cm2 Si 30 keV

5x1014/cm2 Si 30 keV

«SiGeC

B B

T=800ºC15 min, N2

Bor

onD

iffu

sivi

tyE

nhan

cem

ent

(Dox

/DN

2)

84

tion case clear enhancement of the phosphorus diffusion is observed by the deeper

location of the phosphorus. The phosphorus diffusivities for each case were obtained

by fitting simulated diffused phosphorus profiles to experiment with PROPHET using

a spatially constant phosphorus diffusivity. The measured diffusivity is, therefore, a

spatial average of the actual diffusivity, which is not believed to be uniform above the

SiGeC layer (Fig. 4.5). Because phosphorus is believed also to diffuse almost entirely

by an interstitial mechanism [19], the enhanced phosphorus diffusivity is likewise a

measure of the relative silicon interstitial enhancement,

(4.1)

analogous to case of enhanced boron diffusion due to excess interstitial concentra-

tions.

** I

I

D

D

P

measP ≅

85

Figure 4.7 Phosphorus concentration profiles (measured using SIMS) of as-grownall-silicon sample and all-silicon annealed at 897°C for 30 minutes in nitrogen or oxy-gen ambient. Phosphorus OED is evident by the deeper phosphorus edge after oxida-tion.

The phosphorus diffusivity during oxidation is found to be 8 times that of the

average phosphorus diffusivity during the nitrogen anneal in this case agreeing reason-

ably with previous reports (Fig 3.2). Note: the boron diffusivity was enhanced 6 times

by oxidation in the same sample, which agrees with the phosphorus enhancement

within the uncertainty of the measurement. In Fig. 4.8, the phosphorus profiles before

and after oxidation in the all-silicon case is overlaid on that from the sample with a

buried Si0.795Ge0.20C0.005. The annealed carbon profile is overlaid on the phosphorus

10 16

10 17

10 18

10 19

0 0.04 0.08 0.12 0.16

Pho

spho

rus

Con

cent

rati

on[c

m-3

]

Depth [µm]

897°C, 30 min

As-GrownN2

O2

86

profiles for reference. After oxidation the phosphorus edge in the silicon sample

(dashed line) has clearly diffused deeper into the silicon layer than that in the sample

with a buried SiGeC layer (solid line), despite the higher initial phosphorus concentra-

tion in the as-grown sample with the buried SiGeC layer (solid line). A four times

Figure 4.8 Phosphorus concentration profiles (measured using SIMS) before andafter annealing the samples at 897°C for 30 minutes in oxygen ambient with (solidline) or without (dashed line) a buried SiGeC layer. The carbon concentration (dottedline) of the buried SiGeC layer after annealing is overlaid for reference.

reduction of phosphorus diffusivity is observed in the sample containing the buried

SiGeC, even though the SiGeC layer is 1500-2000Å below the phosphorus moving

edge. A summary of the measured diffusivities after oxidation relative to the mea-

10 17

10 18

10 19

10 20

10 21

0 0.1 0.2 0.3

Ato

mic

Con

cent

rati

on[c

m-3

]

Depth [µm]

All-Si

SiGeC

C

897ºC,30 min, O2

As-Grown

P

87

sured diffusivities after the nitrogen anneal are shown in Fig. 4.9 for the silicon case

and two carbon concentrations of 0.05% and 0.5%. The phosphorus OED above the

SiGeC layer decreases with increasing carbon level as was observed for the boron

marker layer below the SiGeC layer, Fig. 4.4 and 4.9. The interstitial concentration

above and below the SiGeC layer clearly depends on the efficiency of the SiGeC layer

to react with the injected interstitials and presumably the increased number of carbon

react with an increased number of interstitials, which depletes the surface region of

excess interstitials, explaining the reduced phosphorus OED above the SiGeC layer,

Fig. 4.5. However, the exact interstitial profile can not be deduced from the phospho-

rus edge because for short times it samples only one point above the SiGeC layer and

at longer times it is difficult to accurately model the diffusion with a single diffusivity

since the interstitial concentration is presumably varying as a function of depth, as

postulated in Fig. 4.5. The following chapter examines the interstitial concentration

above the SiGeC layer in more detail.

88

Figure 4.9 Relative phosphorus diffusivity dependence on carbon levels in buriedSi(GeC) layers. Average diffusivities during annealing at 897°C for 30 minutes inoxygen ambient is compared to that after annealing in nitrogen.

4.4 Carbon profile after annealing

Because carbon in some defects states is electrically active, it is important to

determine where the carbon is after processing. Furthermore, carbon is reported to

diffuse quickly in silicon [9], making it difficult to contain in a confined region. The

carbon diffusion after oxidation in samples with the highest initial carbon levels in the

SiGeC layer (Fig. 4.10), have nearly the same concentrations as those after annealing

in nitrogen for the same times. Diffusion of carbon at low concentrations in silicon,

less than 1018 cm-3, is usually described well by Fick’s 2nd law [9]. However the

0

2

4

6

8

10

SiGeC0.5% C

Si SiGeC0.05% C

897°C, 30 minP

hosp

horu

sD

iffu

sivi

ty

Enh

ance

men

t(D

Ox/

DN

2)

No OED

89

unusually non-gaussian out-diffusion of carbon showing long tails of low concentra-

tion carbon diffusing hundreds of nanometers surrounding by a nearly stationary high

concentration carbon rich layer has led some researchers to invoke immobile clusters

of Cs-Ci or silicon-carbide precipitation to explain both the interstitial sinking capacity

of carbon as well as the unusual diffusion profiles [2, 20]. The formation of these two

silicon-carbon species can remove silicon interstitials from the local population by

trapping and immobilizing silicon interstitials as Cs-Ci clusters [9]

(mobile) (4.2)

(immobile) (4.3)

or annihilating interstitials through the formation of additional vacancies created by

the silicon-carbide precipitation [21]. The entire carbon profile is described, therefore,

as a large fraction of immobilized carbon clusters or precipitates surrounded by the

out-diffusion of the much smaller concentration of unbound carbon that diffuse via the

carbon’s normal diffusion mechanism (e.g. eq’n. 4.2). It will be shown in the follow-

ing chapters that a more simple reaction pathway sufficiently explains both the sink

effect and the unusual carbon diffusion without invoking a cluster/precipitation reac-

tion.

isCIC →+

isisCCCC −→+

90

Figure 4.10 Carbon concentration profiles (measured using SIMS) of samples thatwere as-grown, or annealed at 897°C for 30 minutes in nitrogen ambient, or annealedat 897°C for 30 min in oxygen ambient.

4.5 Summary

In summary, boron OED and TED in silicon has been completely suppressed by trap-

ping excess interstitials in an overlying Si1-x-yGexCy layer. The boron doped silicon

region remains free of carbon under conditions when boron OED/TED is completely

suppressed, ruling out Bi-Cs as the mechanism responsible for the reduced boron

OED/TED. Due to recombination of interstitials at the surface after implantation and

an as of yet unquantified interstitial injection rate during oxidation the exact relation-

10 16

10 17

10 18

10 19

10 20

10 21

0.2 0.22 0.24 0.26 0.28 0.3

Car

bon

Con

cent

rati

on[c

m-3

]

Depth [µm]

Asgrown

O2

N2

897°C, 30 min

91

ship between substitutional carbon and silicon interstitials remains to be resolved in

the following chapters.

92

4.6 References

[1] L. D. Lanzerotti, J. C. Sturm, E. Stach, R. Hull, T. Buyuklimanli, and C.Magee, IEDM Tech. Digest , pp. 249, 1996.[2] P. A. Stolk, H.-J. Gossmann, D. J. Eaglesham, and J. M. Poate, Mat. Sci. &Eng. , vol. B36, pp. 275-281, 1996.[3] H. Ruecker, B. Heinemann, W. Roepke, R. Kurps, D. Krueger, G. Lippert, andH. J. Osten, “Suppressed diffusion of boron and carbon in carbon-rich silicon,” Appl.Phys. Lett. , vol. 73, pp. 1682, 1998.[4] R. Scholz, U. Goesele, J. Y. Huh, and T. Y. Tan, Appl. Phys. Lett. , vol. 72, pp.2, 1998.[5] A. Mocuta, R. Strong, and D. Greve, presented at Electronic Materials Confer-ence (40th), Charlottesville, VA, 1998.[6] L. D. Lanzerotti, J. C. Sturm, E. Stach, R. Hull, T. Buyuklimanli, and C.Magee, Appl. Phys. Lett. , vol. 70, pp. 23, 1997.[7] P. A. Stolk, H.-J. Gossmann, D. J. Eaglesham, D. C. Jacobson, C. S. Rafferty,G. H. Gilmer, M. Jaraiz, J. M. Poate, H. S. Luftmann, and T. E. Haynes, “Physicalmechanisms of transient enhanced dopant diffusion in ion-implanted silicon,” J. Appl.Phys. , vol. 81, pp. 6031, 1997.[8] H. J. Osten, G. Lippert, P. Gaworzewski, and R. Sorge, Appl. Phys. Lett. , vol.71, pp. 11, 1997.[9] G. Davies and R. C. Newman, “Carbon in monocrystalline Silicon,” in Hand-book on Semiconductors , T. S. Moss, Ed., 1994, pp. 1558.[10] M. Carroll, C.-L. Chang, J. C. Sturm, and T. Buyuklimanli, “Complete sup-pression of boron transient-enhanced diffusion and oxidation-enhanced diffusion insilicon using localized substitutional carbon incorporation,” Appl. Phys. Lett. , vol. 73,pp. 3695, 1998.[11] M. S. Carroll, L. D. Lanzerotti, and J. C. Sturm, “Quantitative measurement ofreduction of boron diffusion by substitutional carbon incorporation,” presented at Dif-fusion Mechanisms in Crystalline Materials, San Fransisco, 1998.[12] M. S. Carroll, J. C. Sturm, and C.-L. Chang, “Quantitative measurement ofreduction of phosphorus diffusion by substitutional carbon incorporation,” presentedat Proceedings of the 1999 MRS Spring Meeting - Symposium on 'Si Front-End Pro-cessing - Physics and Technology of Dopant-Defect Interactions', San Fransisco, CA,USA, 1999.[13] J. C. Sturm, P. V. Schwartz, E. J. Prinz, and H. Manoharan, J. Vac. Sci. Tech. B, vol. 9, pp. 2011, 1991.[14] Fahey, P. B. Griffin, and J. D. Plummer, Rev. Mod. Phys. , vol. 61, pp. 289,1989.

93

[15] M. R. Pinto, D. M. Boulin, C. S. Rafferty, R. K. Smith, W. M. Coughran, I. C.Kizilyalli, and M. J. Thoma, Tech. Digest IEDM , pp. 923, 1992.[16] N. Moriya, L. C. Feldman, S. W. Downeym, C. A. King, and A. B. Emerson,“Interfacial Segregation in Strained Heterostructures: Boron in Si0.8 Ge0.2 /Si,” Phys.

Rev. Lett. , vol. 75, pp. 1981, 1995.[17] S. M. Hu, D. C. Ahlgren, P. A. Ronsheim, and J. O. Chu, Phys. Rev. Lett. , vol.67, pp. 1450-53, 1991.[18] W. Fang, P. Griffin, and J. Plummer, presented at Materials Research Sympo-sium, 1995.[19] A. Ural, P. B. Griffin, and J. D. Plummer, “Fractional contributions of micro-scopic diffusion mechanisms for common dopants and self-diffusion in silicon,” J.Appl. Phys. , vol. 85, pp. 6440, 1999.[20] J. W. Strane, H. J. Stein, S. R. Lee, S. T. Picraux, J. K. Watanabe, and J. W.Mayer, J. Appl. Phys. , vol. 76, pp. 3656, 1994.[21] W. J. Taylor, W. Y. Tan, and U. Goesele, Appl. Phys. Lett. , vol. 62, pp. 2336,1995.

94

Chapter 5

Quantitative Measurement of the Surface Silicon Interstitial Boundary Condition

and Silicon Interstitial Injection into Silicon during Oxidation

5.1 Introduction

In the previous chapter, it was found that if enough substitutional carbon was added to

a SiGe layer, then the resulting SiGeC layer could act to completely prohibit intersti-

tials from passing beyond the layer. The SiGeC layers, therefore, can insulate dopant

layers below the SiGeC from TED and OED effects. Furthermore, even when the

SiGeC layer was located 250 nm below the diffusing species it reduces the OED

enhancement below the pure silicon case, demonstrating that the carbon acts as an

effective sink for injected interstitials reducing the total number of interstitials in the

surface region. However, to quantify the carbon-interstitial relationship it is necessary

to measure both the number of injected interstitials and the number of substitutional

carbon affected. Because the number of interstitials injected after typical processes

like ion implantation or oxidation are not well characterized, it is necessary to first

quantify the interstitial injection. We choose to examine oxidation. Although intersti-

tial formation due to ion implantation is a well studied system due to its importance to

transistor fabrication, the interstitial formation and injection processes are known to be

complex and difficult to predict because of incomplete knowledge about most key pro-

95

cesses, including interstitial-vacancy (Frenkel pair) formation by the initial implant,

interstitial self-cluster condensation and evaporation kinetics, surface recombination

and generation rates, Frenkel pair post implant recombination and interstitial annihila-

tion due to amorphization and subsequent recrystallization. In this chapter, using

modified boron marker layer test structures, the interstitial profile above the SiGeC

layer is mapped out. From this profile we are able to quantify the interstitial boundary

condition during oxidation and thereby determine the number of injected interstitials

from the surface. The following chapter will use this chapter’s results to quantify the

loss of substitutional carbon for a known interstitial injection rate.

5.2 Interstitial profile above buried SiGeC layers

5.2.1 Test structures

Test structures were grown to measure the local boron diffusivity throughout the sur-

face region of samples containing zero (sample A), one (sample B), or two (sample C)

buried Si0.795Ge0.2C0.005 layers (Fig 5.1(a), (b), and (c) respectively). The test struc-

tures were epitaxial layers grown on silicon substrates using rapid thermal chemical

vapor deposition (RTCVD) at temperatures between 625°C and 750°C using dichlo-

rosilane, germane, and methylsilane as the silicon, germanium and carbon sources

respectively [1]. Each test structure was grown on a p-type Czochralski (CZ) (100)

silicon wafer and was grown on both the top and bottom surface of the silicon wafer

because of the reactor geometry. The three different test structures were grown with

96

four 25 nm thick boron marker layers that had peak concentrations of 4-9x1018 cm-3

centered below the surface at 150, 450, 600, and 900 nm depths. Sample B was grown

with one 20 nm thick Si0.795Ge0.2C0.005 layer centered at 675 nm below the surface;

sample C was grown with two 20 nm thick Si0.795Ge0.2C0.005 layers centered at 300,

and 675 nm below the surface. Because of non-uniformity across the wafer surface

the depths of the boron layers differed from the nominal values, unintentionally, as

much as 30% from sample to sample.

Figure 5.1. Schematic diagram of the test structures A, B, and C ((a), (b), and (c)respectively) used in this experiment.

All test structures were cleaved and annealed in an oxygen or nitrogen ambient

for various times between 30 and 960 minutes at 750°C or 850°C, and the resulting

boron, carbon, germanium and oxygen profiles were obtained using secondary ion

mass spectrometry (SIMS) done at Evans East in East Windsor, NJ. Samples were

(a) (b) (c)

25 nm, Si0.795Ge0.2C0.005

25 nm, 1019 cm-3 boron

25 nm, Si0.795Ge0.2C0.005

25 nm, 1019 cm-3 boron

25 nm, 1019 cm-3 boron

25 nm, 1019 cm-3 boron

135 nm, i-Si

135 nm, i-Si

135 nm, i-Si

50 nm, i-Si

100 nm, i-Si

150 nm, i-Si25 nm, Si0.795Ge0.2C0.005

25 nm, 1019 cm-3 boron

250 nm, i-Si

25 nm, 1019 cm-3 boron

25 nm, 1019 cm-3 boron

25 nm, 1019 cm-3 boron

135 nm, i-Si

50 nm, i-Si

100 nm, i-Si

150 nm, i-Si275 nm, i-Si

25 nm, 1019 cm-3 boron

250 nm, i-Si

25 nm, 1019 cm-3 boron

25 nm, 1019 cm-3 boron

25 nm, 1019 cm-3 boron

135 nm, i-Si

150 nm, i-Si

Si-substrate Si-substrate Si-substrate

97

sputtered using 2 keV Cs+ ions, and depths were determined using standard profilom-

etry of the sputtered craters leading to a 5% uncertainty in depths and a 20% uncer-

tainty in boron concentrations. The oxide growth rates measured by ellipsometry were

0.33 Å/min and 0.91 Å/min at 750°C and 850°C respectively, in agreement with previ-

ous reports of thin silicon oxide films [2]. No systematic difference was observed

between the oxidation rate of silicon surfaces containing buried SiGeC layers and

those without buried SiGeC layers. The most silicon consumed was 130Å. This was

ignored in subsequent analysis where the depths from the surface were required.

Boron profiles of the pure silicon structure (sample A) after annealing at 850°C

for 30 minutes in oxygen or nitrogen ambient are noticeably broader than the as-grown

case (Fig. 5.2(a)). Moreover, the boron profiles in sample A after annealing

Figure 5.2. Boron and carbon concentration profiles from SIMS of as-grown (dashedline, labeled B and C above the respective profiles) and boron concentrations for sam-ples annealed for 30 minutes at 850°C in either oxygen (solid line) or nitrogen (solid-dashed line) ambients for structures A, B, and C ((a), (b), and (c)) respectively.

1017

1018

1019

1020

1021

0 0.1 0.2 0.3 0.4 0.5

Con

cent

rati

on[c

m-3

]

Depth [µm]

B B

C

1017

1018

1019

1020

1021

0 0.2 0.4 0.6 0.8 1

Con

cent

rati

on[c

m-3

]

Depth [µm]

C

B BB B

1017

1018

1019

1020

1021

0 0.2 0.4 0.6 0.8 1

Con

cent

rati

on[c

m-3

]

Depth [µm]

B B B B

(a) (b) (c)

As-Grown

O 2

N2

As-Grown

O2

N2

As-Grown

O 2

N2

98

in oxygen ambient are clearly broader, at all depths, than those after nitrogen anneal

for the same time and temperature, indicative of the well-documented oxygen

enhanced diffusion effect [3]. Boron profiles in the two samples containing

Si0.795Ge0.2C0.005 layers before and after annealing in the identical conditions as in

sample A (Fig. 5.2(b) and (c)) show different behavior above and below the buried

Figure 5.3. Boron and carbon concentration profiles from SIMS of as-grown (dashedline, labeled B and C above the respective profiles) and boron concentrations for sam-ples annealed for 120 minutes at 850°C in either oxygen (solid line) or nitrogen (solid-dashed line) ambients for structures A, B, and C ((3a), (3b), and (3c)) respectively.

Si0.795Ge0.2C0.005 layers. Boron profiles below the Si0.795Ge0.2C0.005 layers after 30

minutes of oxidation are identical to those after nitrogen anneal. As discussed in the

previous chapter, the carbon layer prohibits interaction between the injected intersti-

tials from the surface region and the boron below the carbon layer for this oxidation

condition [4]. Boron profiles after oxidation above the Si0.795Ge0.2C0.005 layers are,

however, broader than their respective counterparts annealed in nitrogen ambient. The

1017

1018

1019

1020

1021

0 0.2 0.4 0.6 0.8 1

Con

cent

rati

on[c

m-3

]

Depth [µm]

BB B B

1017

1018

1019

1020

1021

0 0.1 0.2 0.3 0.4 0.5

Con

cent

rati

on[c

m-3

]

Depth [µm]

BB

C

1017

1018

1019

1020

1021

0 0.2 0.4 0.6 0.8 1

Con

cent

rati

on[c

m-3

]

Depth [µm]

BB

C

BB

(a) (b) (c)

As-Grown

O2

N2

As-Grown

O2

N2

As-Grown

O2

N2

99

differences in boron profile widths after oxidation versus nitrogen anneals, from

marker to marker, are not however uniform. The profiles are clearly broader the

Figure 5.4. Boron and carbon concentration profiles, from SIMS of as-grown (dashedline, labeled B and C above the respective profiles) and boron concentrations for sam-ples annealed for 240 minutes at 750°C in either oxygen (solid line) or nitrogen (solid-dashed line) ambients for structures A, B, and C ((4a), (4b), and (4c)) respectively.

nearer the boron marker is to the surface (Fig. 5.2(b)). The depth dependent diffusivity

indicates a gradient in the interstitial concentration in the surface region. Boron con-

centration profiles after annealing in nitrogen or oxygen ambient were obtained for a

number of different times and temperatures and are shown overlain on their respective

as-grown profiles for 850°C after 120 minutes (Fig. 5.3), and for 750°C after 240, and

960 minutes (Fig. 5.4, and Fig. 5.5 respectively).

1017

1018

1019

1020

1021

0 0.2 0.4 0.6 0.8 1C

once

ntra

tion

[cm

-3]

Depth [µm]

BB

C

BB

1017

1018

1019

1020

1021

0 0.2 0.4 0.6 0.8 1

Con

cent

ratio

n[c

m-3

]

Depth [µm]

BB B B

1017

1018

1019

1020

1021

0 0.1 0.2 0.3 0.4 0.5

Con

cent

ratio

n[c

m-3

]

Depth [µm]

B B

C

(a) (b) (c)

As-Grown

O2

N2

As-Grown

O2

N2

As-Grown

O2

N2

100

Figure 5.5. Boron and carbon concentration profiles from SIMS of as-grown andboron concentrations for samples annealed for 960 minutes at 750°C in either oxygenor nitrogen ambients for structures A, B, and C ((5a), (5b), and (5c)) respectively.

5.2.2 Extraction of Diffusivity Enhancement and Interstitial Profile

Average local boron diffusivities during annealing for each individual boron

marker were extracted for each peak of each sample by using PROPHET [5] to numer-

ically solve for the boron profiles after annealing using the as-grown boron concentra-

tions obtained by SIMS as the initial conditions, as described in the previous chapter.

Because the local boron diffusivity depends on the local interstitial concentration as:

(5.1)

the interstitial supersaturation may be mapped throughout the surface region. In

this case, is the literature value for the extrinsic diffusivity of boron in silicon,

eq’n. 3.9, and is the intrinsic self-interstitial concentration in the silicon. The boron

1017

1018

1019

1020

1021

0 0.2 0.4 0.6 0.8 1

Con

cent

ratio

n[c

m-3

]

Depth [µm]

BB B B

1017

1018

1019

1020

1021

0 0.2 0.4 0.6 0.8 1

Con

cent

ratio

n[c

m-3

]

Depth [µm]

BB

C

BB

1017

1018

1019

1020

1021

0 0.1 0.2 0.3 0.4 0.5

Con

cent

ratio

n[c

m-3

]

Depth [µm]

B B

C

(a) (b) (c)

As-Grown

O2

N2

As-Grown

O2

N2

As-Grown

O2

N2

** I

I

D

D

B

measB ≅

*I

I

*B

D

*I

101

diffusivity for the entire depth of the sample was adjusted for each marker layer by

varying a single parameter to fit the annealed boron profile of each marker. The fitting

parameter was the ratio of the interstitial concentration to the intrinsic interstitial con-

centration at the boron marker location, . The interstitial enhancement for each

marker during each oxidation for each time and temperature for all the samples

obtained this way (see end of paragraph) are shown in figures 5.6 (750°C) and 5.7

(850°C). The relative boron diffusivity (i.e. the ratio) measured during nitrogen

anneals after all times examined for sample A (pure silicon) were unity within the

uncertainty of the measurement, and therefore agreed with those previously reported

for the intrinsic boron diffusion in silicon (i.e. the boron diffusion observed during

annealing in inert ambient) [6]. The boron diffusivities during nitrogen anneal in the

samples containing SiGeC layers were, however, slightly higher than that in the pure

silicon sample by a factor of 2-3. The reason for this will be discussed in chapter 6.

All enhanced boron diffusivities reported in this chapter are compared to the literature

value for the intrinsic boron diffusivity. Note that all boron diffusivities plotted at the

SiGeC layer depths are shown in Fig. 5.6 and 5.7 as approximately zero. These values

were taken from the results of chapter two, in which it was found that the boron diffu-

sivity was much smaller than the intrinsic boron diffusivity in silicon (even in cases of

relatively high interstitial injection). The diffusivities shown for the sample C anneals

*I

I

*I

I

102

of 60 or 120 min at 850°C and 960 min at 750°C were obtained by a second method

explained in the following section, which was necessary to accommodate for the

asymmetric boron profile.

Figure 5.6. Fitted boron diffusivity enhancements for all the samples and their markerdepths are shown for oxidations at 750°C for samples A and B after oxidation of (a)240 minutes and (b) 960 minutes.

The diffusivity enhancement for all the pure silicon samples (A) is found to be

fairly uniform throughout the depth of the samples for both temperatures and all times,

e.g. Fig. 5.6(a). This is consistent with previous reports of oxidation-enhanced-diffu-

sion that demonstrate that the interstitial point defects, which enhance the boron diffu-

sivity, may diffuse relatively long distances from the surface in a short time [7]. The

average diffusivity enhancement in silicon at 750°C and 850°C was found to be

approximately 20.5 and 14.8 for the all silicon samples, consistent with previous

reports of the oxidation enhanced diffusion of boron [8].

(a) (b)

0

5

10

15

20

25

30

35

0 200 400 600 800 1000Boron Marker Depth [nm]

0

5

10

15

20

25

30

35

0 200 400 600 800 1000Boron Marker Depth [nm]

Inte

rsti

tialE

nhan

cem

ent[

I/I* ]

Inte

rstit

ialE

nhan

cem

e nt[

I/I* ]

750°C240 min

750°C960 min

(B) (B)

(A)

(A)

103

Figure 5.7. Fitted boron diffusivity enhancements for all the samples and their markerdepths are shown for oxidations at 850°C for samples A, B and C after oxidation of (a)30 minutes (b) 60 minutes (no values are available for sample A for this condition),and (c) 120 minutes.

Comparing the diffusivity enhancements from one oxidation time to another in

a single sample shows that the enhancements remain relatively constant for all oxida-

tion times. This indicates that the interstitial concentration is in a steady-state condi-

tion for these times. Furthermore, as shown in Fig.’s 5.6 and 5.7, in the samples with

buried SiGeC layers, the ratio during oxidation decays from the surface enhance-

ment to approximately zero at the SiGeC layer and is well approximated by a linear

profile versus depth.

The linear fits of the relative interstitial concentration can also be extrapolated

to the surface of the silicon/oxide interface from the interstitial profiles in figure 5.6

and 5.7. The extrapolated relative surface interstitial concentrations for each tempera-

ture and SiGeC depths, averaging the values for the different oxidation times, are dis-

0

5

10

15

0 200 400 600 800 1000Boron Marker Depth [nm]

Inte

rsti

tialE

nhan

cem

ent[

I/I*

]

(a)

850°C30 min

(B)(C)

0

5

10

15

0 200 400 600 800 1000Boron Marker Depth [nm]

(b)

850°C60 min

Inte

rsti

tialE

nhan

cem

ent[

I/I*

]

(B)(C)

0

5

10

15

0 200 400 600 800 1000Boron Marker Depth [nm]

(c)

850°C120 min

Inte

rstit

ialE

nhan

cem

ent[

I/I*

]

(B)(C)

(A)(A)

*I

I

104

played versus the depth of the SiGeC layers in figure 5.8. For comparison the surface

concentrations obtained for sample A (no SiGeC layer) are shown on the same figure

at a depth of 485 µm (the average depth of the back surface from the top surface). The

uncertainty of the extrapolated interstitial surface concentrations, resulting from the

uncertainty in the best linear fits, is indicated by the error bars in figure 5.8. The sur-

face interstitial concentrations measured for each sample are within 20% of the aver-

age surface concentrations (averaged over all samples for each temperature), ~25 and

~12.7 for 750°C and 850°C, respectively. Possible effects of interstitials diffusing to

the top region from the oxidizing back surface, that might affect the measured boron

diffusivity enhancements, are neglected in cases B and C, because SiGeC layers were

grown on both the top and bottom surfaces. Since the SiGeC layer on the back surface

will react with injected interstitials from that surface first and the carbon content has

been demonstrated to be enough to sink all injected interstitials, no additional intersti-

tials are expected from the rear surface. This conclusion is supported by the lack of

enhanced diffusion of the boron profiles below the SiGeC layer on the top surface. In

case A (all-silicon sample) boron diffusivity enhancements have been previously

found to be completely independent of the proximity of local interstitial sources and

drains like a rear oxidizing surface as close as 20 µm [9, 10], which is consistent with

this work’s observation that the surface concentration is relatively independent of the

105

depth of the interstitial sink, even when the nearby interstitial concentration is signifi-

cantly altered by the presence of the interstitial sink.

Figure 5.8. Relative interstitial super-saturation at the surface of the silicon duringoxidation estimated from the boron diffusivity enhancement profiles versus the SiGeClayer depth. For comparison, the depth of the rear surface is used to represent the puresilicon case located approximately 500 µm from the top surface.

5. 3 Boron Diffusion in a Linear Diffusivity Gradient

The above method for extraction of the boron diffusivity could be used as long as the

boron peaks remained narrow and the ratio of (or boron diffusivity enhancement)

is approximately constant over the region in which the boron is diffusing. However, in

sample C (Si0.795Ge0.2C0.005 layer ~30 nm below the surface), the boron concentration

profile after 120 minutes of oxidation at 850°C and 960 minutes at 750°C are clearly

very asymmetric (Fig. 5.3(c) and 5.5(c)) due to the diffusion in a steep interstitial gra-

5

10

15

20

25

30

35

40

100 1000 104 105 106

750°C

850°C

Sur

face

Inte

rstit

ialC

once

ntra

tion

[I/I

* ]

Depth of SiGeC Layer [nm]

(A)Pure Si

(B)(C)

*I

I

106

dient in the surface region. Near the surface the diffusion coefficient is substantially

higher than near the SiGeC layer. It is no longer appropriate to extract a single diffu-

sivity from the boron profiles in these cases. Therefore, for sample C it was necessary

to simulate the boron diffusion in a diffusivity gradient to reproduce the asymmetric

broadening and estimate the boron diffusivity enhancement. Boron profiles after

annealing were simulated, using PROPHET, with a linearly decaying interstitial

enhancement profile (i.e. boron diffusivity). The interstitial concentration was

assumed pinned at zero at the SiGeC layer, and the interstitial enhancement at the sur-

face ( ) was used as a single adjustable parameter to fit simulated profiles

with the data. Fig. 5.9 shows the excellent fit of the modeling with the data, support-

ing the assumption of a linearly decaying interstitial enhancement profile. (For refer-

ence the carbon profiles before and after oxidation are also shown in figure 5.9 and

also show that the carbon profile has not changed appreciably in the surface region.)

*surf

I

In

surf≡

107

Figure 5.9. (a) As-grown boron and carbon concentration profiles overlain on boronand carbon profiles after 120 minutes oxidation at 850°C and simulated profile (solidline) of the interstitial enhancement after oxidation for 120 minutes at 850°C for sam-ple C (SiGeC layer located 300 nm below the surface).

5.4 Interstitial Injection Rate

5.4.1 Calculation of Interstitial Injection Rate

If the interstitial concentration at the surface, , and the interstitial diffusivity, ,

were known one could easily calculate the diffusion flux of interstitials from the sur-

face for a given SiGeC layer depth, assuming a linear decay of the concentration pro-

file. However, we have measured the relative interstitial enhancement, , and the

1017

1018

1019

1020

1021

0 0.1 0.2 0.3

Con

cent

rati

on[c

m-3

]

Depth [µm]

B

B

C

C

surfI *

ID

*I

I

108

actual intrinsic concentration and diffusivity of silicon self-interstitials is unknown.

Nevertheless we can still calculate the flux because the interstitial transport prod-

uct (68 cm-1s-1 and 1x104 cm-1s-1 for 750°C and 850°C, respectively) has been

measured reliably using metal tracer diffusion [11]. For samples B and C the silicon

interstitial flux, , injected into the silicon may be calculated by,

(5.2)

where, is the depth of the Si0.795Ge0.2C0.005 layer, and is the experimentally

obtained relative interstitial super-saturation at the surface (Fig 5.8). It is assumed that

the interstitial concentration at the SiGeC layer is near zero [4, 12]. Evaluating eq’n.

5.2 for 750°C and 850°C gives:

and

where there is a total 25% uncertainty in the values. Empirically estimating the sur-

face self-interstitial enhancement ratio from Fig. 3.2 as:

**IDI

IJ

x

IDn

x

ID

I

I

dx

dIDJ I

surfI

II ∆=

∆×=−=

****

**

x∆ surfn

1131009.2

)750( −−⋅∆×== scmatomsx

CTJIo (5.3)

1151051.1

)850( −−⋅∆×== scmatomsx

CTJIo (5.4)

Tk

eV

surfBen7.0

31033.9 −×= (5.5)

109

the temperature dependence of the self-interstitial injection may also be expressed as:

The total number of injected interstitials for samples B and C can be calculated

by integrating the interstitial flux injected during oxidation (eq’n. 5.3 and 5.4) over the

oxidation time (960 or 120 minutes at 750°C or 850°C, respectively), Fig. 5.10. In the

case of pure silicon, the interstitial diffusion length, assuming the self-interstitial diffu-

sivities from Bracht’s work, are of the order of the wafer dimension for the longest

oxidation times, i.e. 792 µm or 685 µm for 960 or 120 minutes at 750°C or 850°C,

respectively [11]. An order of magnitude estimate of the number of injected

interstitials into sample A (pure silicon) can be made, therefore, by assuming that the

interstitial concentration has reached a uniform interstitial concentration across the

wafer depth equal to the surface concentration by these oxidation times [9]. The total

number of injected interstitials, therefore, is half the total increase of interstitials in the

entire wafer, i.e. each surface contributes an equal number of interstitials:

where is the width of the wafer (~ 485 µm), is the total number of injected inter-

stitials, and is the literature value for the less well established silicon intrinsic con-

centration (6.2x108 cm-3 or 1.54x1010 cm-3 for 750°C and 850°C, respectively [11]).

11

25.4241038.1),( −−

−

⋅∆×=∆ scmatomsex

xTJ Tk

eV

IB (5.6)

2

)1( * wInQ surf

I

⋅⋅−= (5.7)

wI

Q

*I

110

Figure 5.10. The total interstitial atoms integrated over time injected into the siliconafter 960 minutes at 750°C or 120 minutes at 850°C of oxidation versus the depth ofthe interstitial sink, i.e. the SiGeC layer or the bottom surface. The total interstitialareal densities were calculated using equation 2 or 3 and the measured surface super-saturation and SiGeC layer depths.

Samples B and C (those with a buried SiGeC layer) show a significant increase

in interstitial flux into the silicon bulk compared to sample A (pure silicon) in figure

5.10. The interstitial flux into the silicon bulk is governed by simple diffusion and is

therefore determined entirely by the interstitial gradient. The calculated increase of

interstitial injection due to the proximity of the SiGeC layer is, therefore, because the

SiGeC layer acts as a local sink for injected interstitials reducing the interstitial con-

centration in the region directly below the surface, which in turn draws more intersti-

108

109

1010

1011

1012

1013

1014

100 1000 104 105 106

850°C

750°C

Tot

alSe

lf-I

nter

stit

ials

Inje

cted

[ato

ms

cm-3

]

Interstitial Sink Depth [nm]

PureSi

960 min

120 min

111

tials from the surface into the bulk. When there is no SiGeC layer present the

interstitial concentration rapidly increases and approaches the surface concentration,

which reduces the interstitial gradient at the surface and therefore reduces the intersti-

tial flux into the silicon from the surface (sample A, pure silicon).

5.4.2 Discussion and Comparison to Other Work

The observation that the surface super-saturation of interstitials remains

unchanged (no more than 20% variation from the average) despite an increase of the

total number of injected silicon interstitials qualitatively agrees with the proposed oxi-

dation model by Dunham [13-15]. This model predicts that the interstitial concentra-

tion at the silicon surface is pinned by a large reservoir of silicon interstitials that form

and reside at the oxide/silicon interface above the silicon surface. Briefly, Dunham et

al. propose that in the linear oxidation regime a constant interstitial concentration at

the oxide/silicon interface results from the detailed balance between the generation of

interstitial silicon atoms from the oxidation reaction and primarily the loss of intersti-

tial silicon atoms to diffusion from the interface towards the oxide surface. Presum-

ably this diffusive flux is independent of the oxide thickness, because all the interstitial

silicon in the oxide (a relatively small number compared to the number of oxygen

forming the oxide) react completely with the oxygen species diffusing in towards the

silicon surface in a distance much smaller than the oxide thickness. Naturally, there is

a transient time expected for the steady-state condition and the initial oxide formation.

112

The interstitial concentration measured on the silicon side is predicted, then to be a

small fraction of that on the oxide side, due to a segregation effect driven by the ener-

getically favorable difference in strain. These combined processes, therefore, create

an effective reservoir of interstitials on the oxide side, which pin the interstitial con-

centration at the silicon surface.

This is the first report, of which the author is aware, that shows the boundary

condition remains constant, at 850°C, for interstitial injection rates ranging over 4

orders of magnitude. This demonstrates the stiffness of the surface boundary condi-

tion during oxidation and suggests that the surface concentration of interstitials in the

silicon is determined by an interstitial reservoir at the silicon/oxide interface created

by the oxidation.

Recently, the integrated interstitial flux over time due to oxidation was also

measured by monitoring the increase in size of type II loop defects located 110 nm

below the surface of an oxidized sample [16]. Because loops grow by collecting inter-

stitials, from the number of loops and their growth rate (observed by TEM) one can

directly calculate the number of interstitials consumed. The density of the loops has

been determined to be high enough to capture most of the injected interstitials, there-

fore, their growth is a measure of the total number of injected interstitials. No knowl-

edge of diffusion coefficients or interstitial enhancement values is required. From

Eq’n (5.2), assuming that the interstitial profile decays linearly to zero at the intersti-

tial sink (i.e. the loop defects), we may expect that the flux of interstitials into the sam-

113

ple with buried loop defects, during oxidation, is constant through out the oxidation

time and depends inversely on the depth of the loop defects. To compare our work

with reference [16] for different oxidation times and sink depth, the total number of

interstitials reported by the growth of loop defects is divided by the reported oxidation

time (60 and 120 minutes) to obtain the constant flux during oxidation. The calculated

flux for the loop defect experiment is compared to those calculated for the two sam-

ples B and C after 60 and 120 minutes of oxidation for their respective inverse depths.

As can be seen in figure 5.11, the interstitial fluxes are proportional to the inverse

depths of the interstitial sinks and therefore are well fit by a line for both times. The

loop measurement of the total integrated interstitial injection agrees well with the

boron marker measurements of interstitial injection during oxidation. Furthermore,

the observation that the interstitial injection depends inversely on the depth of the bur-

ied interstitial sink is also in agreement with the observed dependence of loop defect

growth for loops located at varying depths from the surface after oxidation [17].

114

Figure 5.11. The total number of injected silicon interstitials after oxidation measuredusing type II loop defects [16] divided by the oxidation time (i.e. constant interstitialflux during oxidation) compared to the interstitial flux measured by the boron markermethod (this work) is compared for their respective inverse depths. The linear fits forthe two oxidation times show the flux is proportional to the inverse depth.

Although both phosphorus and boron are believed to diffuse almost entirely by

an interstitial(cy) mechanism [18, 19] the boron diffusion enhancements during oxi-

dation are often slightly lower (~30% at 850°C) than that of the phosphorus, [15, 20].

A possible reason for this is boron cluster effects [21], which could temporarily trap

mobile boron and reduce the effective diffusivity causing an underestimate of the true

interstitial injection by the method presented in this work. However, the good agree-

ment between the measurements in this work and that from the growth of loop defects,

0

1 1010

2 1010

0 0.005 0.01

60 min120 min

Tot

alIn

ters

titi

als/

Tim

e[a

tom

scm

-2s-1

]

Inverse Depth of Interstitial Sink [nm-1]

(B)

(C)

Skarlatos et al.

115

which is independent of boron clustering effects, indicates that there are no significant

reductions in boron diffusivity in our work leading to an appreciable underestimate of

interstitial injection. Presumably, the interstitial concentration is still too small to

observe the boron clustering that is observed after implantation. No comparison was

available for 750°C, which may be because the number of injected interstitials calcu-

lated from our data is far below the currently reported resolution of the loop defect

method of 2-3x1013 atoms/cm2. This demonstrates the sensitivity of our approach for

measuring the total number of injected interstitials at very low injection rates limited

primarily by the ability to measure small diffusion lengths (~ 1-2 nm by SIMS) neces-

sary to establish the interstitial profile.

5.5 Summary

The average boron diffusivity during oxidation above Si0.795Ge0.2C0.005 layers has

been used to map the profile of interstitials injected into silicon during oxidation, and

also to quantify the total number of interstitials injected into the silicon substrate. The

interstitial super-saturation concentration at the surface for 750°C and 850°C has been

determined and the average interstitial super-saturation concentration is found to

depend weakly, if at all, on the rate of interstitial injection into the bulk silicon, despite

increasing the total number of injected interstitials by 4 orders of magnitude. In the

following chapter we will quantify the loss of substitutional carbon in similar condi-

tions as these for which we know the interstitial injection rate.

116

5.6 References

[1] J. C. Sturm, P. V. Schwartz, E. J. Prinz, and H. Manoharan, J. Vac. Sci. Tech. B, vol. 9, pp. 2011, 1991.[2] E. A. Irene, J. Electrochem. Soc. , vol. 125, pp. 1708, 1978.[3] D. A. Antoniadis and I. Moskowitz, J. Appl. Phys. , vol. 53, pp. 6788, 1982.[4] M. Carroll, C.-L. Chang, J. C. Sturm, and T. Buyuklimanli, “Complete sup-pression of boron transient-enhanced diffusion and oxidation-enhanced diffusion insilicon using localized substitutional carbon incorporation,” Appl. Phys. Lett. , vol. 73,pp. 3695, 1998.[5] M. R. Pinto, D. M. Boulin, C. S. Rafferty, R. K. Smith, W. M. Coughran, I. C.Kizilyalli, and M. J. Thoma, Tech. Digest IEDM , pp. 923, 1992.[6] R. B. Fair, "Processing Technologies" , vol. Supplement 2B. New York: Aca-demic, 1981.[7] H.-J. Gossmann, C. S. Rafferty, H. S. Luftmann, F. C. Unterwald, T. Boone,and J. M. Poate, “Oxidation enhanced diffusion in Si B-doping superlattices and Siself-interstitial diffusivities,” Appl. Phys. Lett. , vol. 63, pp. 639, 1993.[8] P. Kuo, J. L. Hoyt, J. F. Gibbons, J. E. Turner, and D. Lefforge, “Effects of Sithermal oxidation on B diffusion in Si and strained SiGe layers,” Appl. Phys. Lett. ,vol. 67, pp. 706, 1995.[9] S. T. Ahn, J. D. Shott, and W. A. Tiller, presented at Extended Abstracts of theElectrochemical Society Meeting, 1986.[10] Fahey, P. B. Griffin, and J. D. Plummer, Rev. Mod. Phys. , vol. 61, pp. 289,1989.[11] H. Bracht, N. A. Stolwijk, and H. Mehrer, Phys. Rev. B. , vol. 52, pp. 16542,1995.[12] R. F. Scholz, P. Werner, U. Goesele, and T. Y. Tan, Appl. Phys. Lett. , vol. 74,pp. 392, 1999.[13] S. T. Dunham and J. D. Plummer, “Point-Defect Generation during Oxidationof Silicon in Dry Oxygen Part 2: Comparison to Experiment,” J. Appl. Phys. , vol. 59,pp. 2551, 1986.[14] S. T. Dunham and J. D. Plummer, “Point-Defect Generation during Oxidationof Silicon in Dry Oxygen Part 1: Theory,” J. Appl. Phys. , vol. 59, pp. 2541, 1986.[15] S. T. Dunham and J. D. Plummer, “Interactions of silicon point defects withSiO2 films,” J. Appl. Phys. , vol. 71, pp. 685, 1992.

[16] D. Skarlatos, M. Omri, A. Claverie, and D. Tsoukalas, J. Electrochem. Soc. ,vol. 146, pp. 2276, 1999.[17] D. Tsoukalas, D. Skarlatos, and J. Stoemenos, J. Appl. Phys. , vol. 87, pp.8380, 2000.

117

[18] A. Ural, P. B. Griffin, and J. D. Plummer, “Fractional contributions of micro-scopic diffusion mechanisms for common dopants and self-diffusion in silicon,” J.Appl. Phys. , vol. 85, pp. 6440, 1999.[19] H.-J. Gossmann, T. E. Haynes, P. A. Stolk, D. C. Jacobson, G. H. Gilmer, J. M.Poate, H. S. Luftman, T. K. Mogi, and M. O. Thompson, Appl. Phys. Lett. , vol. 71,pp. 3862, 1997.[20] Roth and Plummer, J. Electrochem. Soc. , vol. 141, pp. 1074, 1994.[21] P. A. Stolk, H.-J. Gossmann, D. J. Eaglesham, D. C. Jacobson, C. S. Rafferty,G. H. Gilmer, M. Jaraiz, J. M. Poate, H. S. Luftmann, and T. E. Haynes, “Physicalmechanisms of transient enhanced dopant diffusion in ion-implanted silicon,” J. Appl.Phys. , vol. 81, pp. 6031, 1997.

118

Chapter 6

Diffusion Enhanced Carbon Loss from SiGeC Layers due to Oxidation

6.1 Introduction

The work in the previous chapters established that substitutional carbon reacts with

silicon interstitials. However, the potential that the interstitial-carbon product is a

defect (i.e ß-SiC precipitation or carbon clusters [1, 2]) may limit the usefulness of

carbon for controlling boron diffusion in technological applications, which highlights

the importance of quantifying the reaction and identifying its product. In this chapter,

carbon out-diffusion from thin SiGeC layers is examined and is shown to be the domi-

nant mechanism of substitutional carbon loss from SiGeC layers close to the surface.

That is, precipitation is suppressed by the rapid loss of all carbon to the surface and the

surrounding unsaturated silicon. Using the results from chapter 5, i.e. the quantifica-

tion of the interstitial injection rate during oxidation, the number of substitutional car-

bon removed from the SiGeC layer due to one injected interstitial is quantified in this

chapter. This is one of the main contributions of this thesis. Furthermore, a complete

quantitative model of the carbon motion is presented that supports the main conclu-

sions of this thesis.

119

6.2 Substitutional carbon loss after oxidation

Two test structures with 25 nm thick Si0.7865Ge0.21C0.0035 layers capped by 300 nm

(structure A) or 40 nm (structure B) of silicon were grown by rapid thermal chemical

vapor deposition (RTCVD) at temperatures between 625ºC and 750ºC using dichlo-

rosilane, germane, and methylsilane as the silicon, germanium, and carbon sources

respectively [3], Fig. 6.1. Each test structure was grown on a p-type Czochralski (CZ)

<100> silicon wafer. Samples of the as-grown and annealed structures were examined

using secondary ion mass spectrometry (SIMS), which were sputtered using 1-2 keV

Cs+ ions, Fig. 6.2, or O+ ions, Fig 6.3. Depths were determined using standard pro-

filometry of the sputtered craters leading to a 5% uncertainty in depths. A 20%, Fig.

6.2, or 10%, Fig.6.3, relative uncertainty in carbon concentrations and approximately a

2% absolute uncertainty in germanium concentrations were obtained.

120

Figure 6.1 Schematic cross section of structures A and B, (a) and (b) respectively.

The as-grown and oxidized samples of structure B were examined by x-ray dif-

fraction (XRD) using a double-crystal rocking curve geometry around the (004) Bragg

reflection. Simulation of the rocking curves of the as-grown and oxidized samples

were fit to experiment using the substitutional carbon concentration in the SiGe layer

as the fitting parameter. The layer thickness and germanium concentration used in the

simulation were taken from the SIMS measurements. The carbon concentration

extracted from the rocking curve fits agreed well with carbon concentrations obtained

by SIMS indicating that the carbon in the SiGeC layer compensated the strain as if it

was all substitutional, within the uncertainty of the measurement, Fig. 6.4 and 6.5.

40 nm, i-Si

Si-substrate

500 nm, i-Si

25 nm, SiGe(C)

25 nm, SiGe(C)

50 nm, i-Si

15 nm, 1019 cm-3 boron

Si-substrate

50 nm, i-Si

15 nm, 1019 cm-3 boron

50 nm, i-Si

15 nm, 1019 cm-3 boron

50 nm, i-Si

15 nm, 1019 cm-3 boron

50 nm, i-Si

50 nm, i-Si

15 nm, 1019 cm-3 boron

500 nm, i-Si

(a) (b)

121

Figure 6.2 Carbon concentration profiles from samples of structure A with buriedSiGeC layers before and after annealing at 850°C in nitrogen or oxygen ambient for240-960 minutes overlaid on the as-grown carbon profile, (a) and (b) respectively.

The strain compensation relationship between germanium and carbon used

was 12:1 [4]. As-grown and 960 minute nitrogen or oxygen annealed samples of

structure A were also examined for relaxation in the plane parallel to the growth sur-

face by scanning around the (224) Bragg reflection. No relaxation by misfit disloca-

tions was observed. Note: The X-ray rocking curves and fits of sample A and B were

done by Dr. D. Tweet at Sharp Labs, WA and Dr. J. Stangl at the University of Linz,

Austria respectively.

1017

1018

1019

1020

1021

0 0.1 0.2 0.3 0.4 0.5 0.6

Car

bon

Con

cent

ratio

n[c

m-3

]

Depth [µm]

1017

1018

1019

1020

1021

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Car

bon

Con

cent

ratio

n[c

m-3

]

Depth [µm]

As-Grown As-Grown 850°C, O2850°C, N2

240 min

960 min

240 min

960 min

(a) (b)

122

Figure 6.3 Carbon concentration profiles from samples of structure B with buriedSiGeC layers before and after annealing at 850°C in nitrogen or oxygen ambient for30-120 minutes overlaid on the as-grown carbon profile, (a) and (b) respectively.

Carbon profiles from samples of structure A, 300 nm silicon cap, annealed in

nitrogen ambient for 240 or 960 minutes are overlaid on the as-grown carbon profile,

Fig. 6.2 (a). The appearance of carbon tails, indicative of carbon out-diffusion, is

observed after annealing. Carbon is believed to diffuse via an interstitial(cy) mecha-

nism when the carbon concentration is below solid solubility [5, 6],

(6.1)

where Cs is a carbon atom in a substitutional site, I is the silicon interstitial, and Ci is

the mobile interstitial carbon defect. Whether the carbon in the tails is actually Ci or

whether it has temporarily been driven substitutional by the reverse reaction releasing

a silicon self-interstitial will be discussed in the section 6.3. The non-Gaussian carbon

1018

1019

1020

1021

0 0.05 0.1 0.15

Car

bon

Con

cent

ratio

n[c

m-3

]

Depth [µm](a) (b)

850°C, N2 850°C, O2

1018

1019

1020

1021

0 0.05 0.1 0.1

Car

bon

Con

cent

ratio

n[c

m-3

]

Depth [µm]

As-Grown As-Grown

30 min30 min

120 min 120 min

60 min

isCIC →+

123

profiles after nitrogen annealing in this case indicate a suppressed carbon diffusivity in

the carbon supersaturated silicon region. The asymmetry of the carbon profiles after

annealing, i.e. lower carbon concentrations on the surface side of the SiGeC layer,

indicates that there is carbon loss from the sample due to carbon diffusion towards and

out of the top surface. Previous studies have also reported loss of carbon from slightly

carbon enriched (8x1017 cm-3) crystalline silicon out the surface after annealing in

either an oxygen or nitrogen ambient [5].

Figure 6.4 X-ray diffraction rocking curves of samples of structure B before and afterannealing at 850°C in oxygen for 0-120 minutes. Rocking curves measured by J.Stangl at the University of Linz.

0.1

1

10

100

1000

104

105

-3000 -2000 -1000 0 1000

Inte

nsit

y[a

rb.u

nits

]

Delta Theta [arcsec]

0.1

1

10

100

1000

104

105

-3000 -2000 -1000 0 1000

Inte

nsit

y[a

rb.u

nits

]

Delta Theta [arcsec]

0.1

1

10

100

1000

104

105

-3000 -2000 -1000 0 1000

Inte

nsit

y[a

rb.u

nits

]

Delta Theta [arcsec]

As-Grown

30 min

120 min

124

The rate of carbon loss from the SiGeC layer is enhanced by oxidation, Fig. 6.2

(b), and the carbon is reduced well below the as-grown background concentration

(3x1017 cm-3), except two spikes of immobile carbon after 960 minutes of oxidation

located at 300 and 370 nm. The oxide-silicon interface is indicated by the carbon

spike located at a depth of 100 nm in this sample. There is a carbon spike which is ini-

tially small and then grows during oxidation outside the SiGeC layer at 370 nm depth

in the as-grown sample Fig. 6.2 (a) and (b). This was unintentional and it is centered

at the same location as a buried 15 nm wide boron layer (grown to examine DB at that

location). Carbon may be preferentially incorporated on boron doped silicon surfaces

during RTCVD or there may be a contamination issue with the gas source switching.

It is unclear whether the enhanced carbon incorporation is substitutional carbon, there-

fore it is difficult to conclude whether the formation of the immobile carbon at 370 nm

depth after oxidation is catalyzed by the surface incorporation during growth or due to

the presence of boron after growth during oxidation, although no immobile carbon for-

mation was initiated in the surface region where four identical boron markers were

spaced evenly between the surface and the SiGeC layer at depths of 50, 125, 200 and

275 nm. Because the amount of carbon in these immobile layers is small, their effect

on the quantification of carbon loss from the sample and SiGeC layer is ignored for the

rest of the chapter.

125

Figure 6.5 Total carbon found in the SiGeC layer of structure B by SIMS compared tothe total substitutional carbon extracted from fits of the rocking curves obtained byvarying the substitutional carbon content and using the germanium concentrationsobtained from SIMS.

In Fig 6.3 (a) the carbon concentration profiles of structure B (40 nm Si cap) of

the as-grown and annealed for 30 or 120 minutes in nitrogen ambient are shown. The

peak carbon concentration is approximately half the as-grown concentration after 120

minutes of annealing without significant broadening of the carbon profile implying

that when carbon diffuses at all, it leaves the SiGeC layer and the sample entirely. Pre-

sumably the primary mechanism of loss is diffusion to the surface and the carbon tails,

if any, are obscured by the higher carbon detection limits (3x1018 cm-3). The carbon

0

1x10 14

2x10 14

3x10 14

4x10 14

0 30 60 90 120

Car

bon

inSi

Ge

Lay

er[c

m-2

]

Oxidation Time [min]

SIMS

XRD

126

concentration profiles after oxidation of 30 to 120 minutes, Fig. 3 (b), shows a more

rapid decrease of the carbon concentration resulting in no detectable carbon in this

sample after 120 minutes of oxidation. Once at the surface the carbon could react with

the oxygen to form CO2 and leave as a volatile product during oxidation, however, this

fails to explain the carbon loss from the surface during annealing in nitrogen. As dis-

cussed earlier, there is a significant barrier to formation of silicon-carbide precipitation

in bulk silicon because of the unusually high surface energies required to form the pre-

cipitate (section 2.3.2). Another possibility, is that the silicon-carbide preferentially

precipitates at the surface and this reaction consumes the carbon. Quantification of the

carbon concentration at the surface is notoriously difficult due to interference with

adventitious carbon contamination on the top surface that interferes with the SIMS

carbon signal in the top 20 nm. It is not possible, therefore, to determine whether sili-

con-carbide is forming at the surface during the nitrogen anneal from the SIMS mea-

surements. More work is still required to identify the mechanism responsible for the

what the carbon consumption at the sample surface.

Immobile carbon within the SiGeC layer (spike located at 300 nm) is observed

in structure A but not in structure B although the initial germanium and carbon con-

centrations are nearly identical and neither layer was grown with appreciable boron in

them (below SIMS detection limits < 4x1016 cm-3). The formation of immobile car-

bon in the SiGeC layer of structure B is prevented by the rapid carbon out-diffusion to

the surface and surrounding silicon compared to structure A. Previous reports of car-

127

bon precipitation or immobile carbon are typically from much thicker carbon layers

(~100 nm thick) in silicon or SiGeC [1, 2, 7]. In these cases the carbon concentration

in the middle of the layer remains near the as-grown value longer because the carbon

out-diffusion is predominantly at the edges of the layers. Indeed, TEM measurements

of thin SiC layers annealed in nitrogen for similar times as in this study showed no

signs of precipitates [8]. The formation of immobile carbon in the SiGeC layer due to

the presence of boron is not likely. The boron concentration throughout all annealing

times was less than 5x1017 cm-3 within the SiGeC layer, 100 times less than the final

immobilized carbon level, and no immobile carbon or boron was observed in the sur-

face region where the carbon diffused through 4 other identically grown boron layers.

Several key conclusions about the carbon loss from the SiGeC layer of struc-

ture B should be emphasized at this point. The rate of carbon loss depends strongly on

the carbon layer’s proximity to the surface. This can play a critical role in the final

chemical composition of a device structure and therefore should be considered in the

design of any fabrication process using substitutional carbon. Furthermore, the rapid

and complete loss of carbon from the SiGeC layer of sample B after oxidation clearly

indicates that the end-product of the reaction between the injected silicon self-intersti-

tial and the substitutional carbon is mobile carbon. No evidence of carbon clusters or

precipitation is seen in this important case.

128

6.3 Quantification of carbon loss

To quantify the loss of carbon from the SiGeC layer and from the sample, the

total carbon detected by SIMS in the SiGeC layers (circle) or in all of the sample

(square) after annealing in either nitrogen (solid) or oxygen (hollow) is subtracted

from the carbon measured in the as-grown samples for both structures A and B, Fig

6.6 (a) and (b), respectively. The rate of carbon loss from both structures after anneal-

ing in nitrogen ambient is initially much more rapid than at later times. Presumably

this is due to the decreasing concentration gradient of mobile carbon at longer times

after the carbon profile has broadened significantly. The rate of carbon loss from the

two SiGeC layers is faster after oxidation. After 120 minutes there is no more remain-

ing carbon in structure B while for structure A it takes more than 240 minutes to

remove a similar amount of carbon. The amount of carbon lost from structure B is

faster than from structure A after the same oxidation condition. The two different

rates of extra carbon lost is shown in Fig. 6.7 for structure A (squares) or B (circles).

More specifically this represents the carbon lost from the SiGeC layer after oxidation

subtracted from that lost after annealing in nitrogen. Clearly, oxidation removes sig-

nificantly more carbon in structure B than in structure A for the same oxidation time.

This is consistent with the result of the previous chapter, in which it was shown that

interstitial injection rate depends inversely on the silicon cap thickness.

129

Figure 6.6 Summary of total carbon lost from the SiGeC layer (circles) or the entiresample (squares) of structures A and B, (a) and (b) respectively, after annealing ineither oxygen (hollow symbols) or nitrogen (solid symbols) ambient at 850°C. Total

carbon in the as-grown SiGeC layers of structure A and B were ~3.5x1014 cm-2 and

4.5x1014 cm-2, respectively.

Using the model for interstitial injection of the previous chapter, the interstitial

flux into the sample is calculated as:

(6.2)

where, nsurf is the ratio of the interstitial surface concentration to the bulk interstitial

concentration (I/I*), DII* is the interstitial transport product measured by metal tracer

diffusion [9, 10], and is the depth of the SiGeC layer. The total number of injected

interstitials after oxidation is calculated as the time integrated flux and is compared to

(b)(a)

N2

O2

1x1014

2x1014

3x1014

4x1014

0 50 100 150Car

bon

Los

tfro

mL

ayer

orS

ampl

e[c

m-2

]

Anneal Time [min]

1x1014

2x1014

3x1014

4x1014

0 200 400 600 800 1000Anneal Time [min]

N2Sample

N2SiGeC

O2SiGeC

O2Sample

Car

bon

Los

tfro

mL

ayer

orS

ampl

e[c

m-2

]

x

IDn

dx

dIDJ I

surfII ∆×=−=

*

x∆

130

the extra carbon that leaves the SiGeC layer due to oxidation for structures A (dotted

line) and B (dashed), (Fig. 6.7).

Figure 6.7 Summary of oxidation enhanced carbon loss from the SiGeC layer, i.e. thedifference of carbon lost between oxidation and nitrogen anneals, for structures A(squares) and B (circles). The number of injected interstitial silicon atoms after oxida-tion is also calculated for structures A (dotted line) and B (dashed line) for compari-son.

The calculated number of interstitials injected into structure B is nearly identi-

cal to the extra carbon that diffuse out of the SiGeC layer due to oxidation, Fig. 6.7,

after 30 and 60 minutes annealing (the value for 60 minutes was interpolated from the

nitrogen data in Fig. 6.5 (a)). Note: the calculated number of interstitials injected

relied solely on the validity of eq’n. 2, the DII* product from the literature, and the

I injected (A)

I inj

ecte

d(B

)

(B) (A)1x1014

2x1014

3x1014

0 60 120 180 240 300

Ato

mic

Con

cent

rati

on[c

m-2

]

Anneal Time [min]

131

depth of the layer. There were no adjustable variables available in this calculation.

After 120 minutes the measured oxidation enhancement saturates and diverges from

the calculated interstitial injection because the carbon, by this time, is entirely

removed from the SiGeC layer. To summarize the results for structure B after oxida-

tion, a silicon self-interstitial injected at the surface during oxidation reacts with the

substitutional carbon in the SiGeC layer (previous chapter). The reaction removes

substitutional carbon from the SiGeC layer and the carbon removed from the layer

agrees quantitatively one-to-one with the number of injected interstitials. The end-

product of the reaction leaves no detectable carbon behind in the form of carbon pre-

cipitation or carbon clusters. We conclude therefore that the carbon reaction with the

excess injected self-interstitials is a “kick-out” like diffusion reaction (eq’n. 6.1) and

the “sink” effect is not due to an immobile carbon trap.

Because the interstitial injection rate depends inversely on the depth of the

SiGeC layer (eq’n. 2), the number of interstitials injected into structure A is less than

B. Therefore the oxidation-enhanced carbon diffusion out of the SiGeC is slower (Fig.

6.7) in A (thick case) than in B (thin case). However, the number of extra carbon

atoms removed from the SiGeC layer in structure B due to oxidation is greater than the

number of interstitials injected into structure A. Assuming carbon diffusion is

described by the reactions in eq’n. 1, one interstitial migrating into the SiGeC layer

can be responsible for the production of only a single mobile carbon. However, one

additional interstitial injected at the surface may be responsible for the removal of

132

multiple carbon atoms from the SiGeC layer, if the mobile carbon that leaves the

SiGeC layer, and then goes back to again occupy a substitutional site, this “kicking”

out a silicon self-interstitial atom, in the surrounding silicon near the SiGeC. The

interstitial is then free to migrate back to the SiGeC layer and remove a second carbon

from the SiGeC layer effectively recycling the interstitial (Fig. 6.7 & 6.8). This

increases the ratio of extra carbon removed to injected interstitials above unity. The

amount of recycling will depend on the carbon layer proximity to the surface, because

some mobile carbon will escape the sample via the surface, thus removing the intersti-

tial completely from the sample. As is indeed observed, the ratio of removed carbon

to injected interstitials is 1:1 for the SiGeC layer closest to the surface and only is

greater than one for the deeper SiGeC layer. The observed 1:1 ratio in structure B also

indicates that the carbon does not immediately form immobile clusters with the

injected interstitials in the SiGeC (i.e. Ci-Cs, Ci-I or Ci-Ix), which have been observed

in silicon, for example, after electron radiation [6]. They may eventually result, but

only if the Ci is not rapidly drawn out of the layer or sample first. A second experi-

ment using boron marker layers to monitor the local interstitial concentration was

done to further examine the recycling effect.

133

Figure 6.8 Schematic diagram of the (i) 1:1 carbon loss vs. I injected kick-out mech-anism for the thin Si cap sample and (ii) for the thick Si cap sample in which the recy-cling frequency is higher.

6.4 Double carbon layer experiment

Previously it has been shown that a SiGeC layer efficiently getters interstitials

that migrate to the carbon layer. This effect can be used to effectively insulate the sili-

con below the SiGeC layer from interstitials introduced from the surface. Therefore, a

region of silicon between two SiGeC layers is effectively insulated from interstitials

injected from the surface and the bulk. Furthermore, self-interstitials in a silicon

region between two SiGeC layers will diffuse to the surrounding carbon layers, where

they are annihilated. These interstitials, based on this simple picture, can not be re-

I

Silicon SiliconSiGeC

Cs+I

Ci

Oxide

i) Either mobile Ciescapes sampleremoving theinterstitial (I)

I

Silicon SiliconSiGeC

Cs+I

Ci

Cs+ICs+I

Ci

1st C removed

2nd C removed

……

Oxideii) Or mobile Ci

decays in Si andreleases theinterstitial (I)

Ci

Ci

134

supplied from the surface or bulk regions, therefore, the interior silicon region will

begin to be depleted of self-interstitials. The steady-state concentration will depend on

a detailed balance between the self-interstitial generation rate and the rate at which the

self-interstitials are lost to the surrounding carbon. This novel structure offers a

method to probe the generation rate of silicon self-interstitials. It will be shown that

the generation of self-interstitials due to recycling is much faster than the bulk genera-

tion rate of silicon self-interstitials.

A structure with four boron marker layers separated by two SiGeC layers was

grown to examine the effect on the boron diffusivity in a silicon region isolated from

the bulk and surface supplies of silicon self-interstitials by two surrounding SiGeC

interstitial sinks (Fig. 6.9). The layers were grown as described section 6.2 and

annealed in nitrogen or oxygen ambient at 850ºC for 30-240 minutes. As-grown and

Figure 6.9 Schematic cross section of the double SiGeC layer with boron marker lay-ers to examine the interstitial concentration in silicon between two interstitial sinks.

25 nm, Si0.795Ge0.2C0.005

25 nm, 1019 cm-3 boron

25 nm, Si0.795Ge0.2C0.005

25 nm, 1019 cm-3 boron

25 nm, 1019 cm-3 boron

25 nm, 1019 cm-3 boron

135 nm, i-Si

135 nm, i-Si

135 nm, i-Si

50 nm, i-Si

100 nm, i-Si

150 nm, i-Si

Si-substrate

135

120 minute nitrogen annealed boron and carbon profiles, Fig. 6.10 (a), show broaden-

ing of both the carbon and boron layers between the two SiGeC layers. A single fit

describing the boron diffusivity in both markers in between the SiGeC layers for each

time after annealing in oxygen or nitrogen ambient are compared to those found in the

pure silicon sample grown on the same day and annealed in identical conditions, Fig.

6.10 (b). The boron diffusivity between the SiGeC layers is enhanced in both the oxy-

gen- and nitrogen- annealed cases after 30 minutes nearly identical amounts.

Figure 6.10 (a) boron and carbon concentration profiles from the double buriedSiGeC layer sample before and after annealing at 850°C in nitrogen ambient for 120minutes and (b) the measured diffusivity of the boron between the SiGeC layers com-pared to that found in pure silicon after annealing in oxygen or nitrogen ambient.

The boron diffusivity becomes and remains indistinguishable from the nitrogen

annealed boron diffusivity in the pure silicon samples after 120 minutes. We con-

10 16

10 17

10 18

10 19

10 20

10 21

0 0.2 0.4 0.6 0.8 1

Ato

mic

Con

cent

rati

on[c

m-3

]

Depth [µm]

0

2

4

6

8

10

12

14

0 50 100 150 200 250Bor

onD

iffu

sion

Coe

ffic

ient

[Dm

eas/D

liter

atur

e]

Time [min]

T=850°C

(a) (b)

B

B B

C C30 min

Pure –Si, O2

Pure –Si, N2

SiGeC, O2

SiGeC, N2

136

clude, therefore, that the interstitial concentration between the SiGeC layers is rela-

tively uniform and not depleted by the nearby SiGeC interstitial sinks despite that

most of the interstitials have had time to diffuse to and react with the carbon (a conser-

vative estimate of the diffusion length of the self-interstitial is ~800 nm for the case of

[C] ~ 1018 cm-3 [11]). Because additional interstitials are prohibited from coming

from either the surface or bulk through the SiGeC layer, the silicon interstitials that

diffuse and react with the carbon in the SiGeC layer must somehow be generated in the

region between the SiGeC layers. There remains the question as to why the diffusion

and thus the interstitial concentration is enhanced in the SiGeC sample for short times

during annealing in either nitrogen or oxygen ambient. It will be shown in section 6.5

that the generation mechanism for additional interstitials can be explained by the recy-

cling mechanism.

Frenkel pair recombination is believed to play a significant role in the removal

of point-defects immediately after ion-implantation [12], therefore a possible source

of interstitial generation might be the reverse reaction in the case of interstitial under-

saturation:

(6.3).

Assuming a constant generation rate of interstitials, a fixed vacancy concentration, and

a vacancy-interstitial recombination rate proportional to the product of the interstitial

and vacancy concentrations, a recombination/generation rate of the interstitials can be

defined by a single effective lifetime of the interstitial, :

VISiS +→

effτ

137

(6.4)

where U is the recombination/generation rate. A characteristic diffusion length of the

interstitial,

(6.5)

describes the exponential decay of an interstitial profile during steady-state diffusion.

A minimum bound on the interstitial diffusion length before recombination with a

vacancy for 850ºC is 700 nm, inferred from the lack of exponential decay in the inter-

stitial profiles between the surface and the SiGeC layer during oxidation in the previ-

ous chapter, Fig. 5.7 (c). However, the uniform intrinsic boron diffusivity observed

between the two SiGeC layers indicates that the interstitial concentration returns to the

intrinsic interstitial concentration in a distance smaller than 100 nm, a much shorter

depth than the minimum recombination decay length. This rules out a simple intersti-

tial generation mechanism like that described by eq’n. 4.

As discussed in section 6.3 another possible source of interstitials is from the

reverse reaction of the kick-out reaction. The mobile carbon interstitial diffuses out of

the substitutional carbon layer and decays after an average diffusion lifetime to a sub-

stitutional site, which in turn releases an interstitial. In the following section the car-

bon diffusion was modeled, based on the kick-out mechanism, to illustrate

eff

IIU

τ

*−=

effII DL τ⋅=

138

quantitatively how the kick-out and reverse reaction fit together to reasonably repro-

duce the experimental observations.

6.5 Simulations

As discussed earlier interstitial diffusing dopants, self-interstitials, and carbon are all

reported to diffuse slower in the presence of supersaturated carbon concentrations. An

initial proposal to explain this effect was the local undersaturation of the interstitial

concentration due to carbon complexing, i.e. trapping, the silicon interstitials. How-

ever, this proposal does not explain the diffusion between the SiGeC layers nor does it

explain how a single interstitial may be responsible for the removal of more than a sin-

gle substitutional carbon atom as was discussed in section 6.3. Alternatively, the

undersaturation of interstitials in the carbon rich regions has also been successfully

explained by the rapid reaction of the substitutional carbon with the local silicon inter-

stitials ultimately depleting the region of interstitials faster than can be supplied from

the surrounding region [13, 14]. To test this hypothesis the proposed model for carbon

in silicon was simulated and compared to the experiment in the previous section.

6.5.1 Model of carbon diffusion

Carbon diffusion is proposed to be a result of the parallel “kick-out” (KO) and Frank-

Turnbull (FT) mechanism that both lead to the formation of the mobile interstitial-car-

bon defect when substitutional carbon reacts with silicon point-defects:

139

(6.6)

(6.7)

where V is a silicon vacancy. Although the FT mechanism is energetically unfavor-

able, the amount of carbon diffusion observed in previous experiments could not be

explained only by the KO mechanism [14]. Reports of enhanced antimony and germa-

nium diffusion in the presence of carbon supersaturated silicon [15], which diffuse via

a vacancy-assisted mechanism, have led to the suggestion of this second vacancy pro-

ducing mechanism (FT) to explain the source of the extra carbon diffusion. These two

reactions lead to the following linked partial differential equations that describe the

carbon and point defect reactions and subsequent diffusion:

(6.8)

(6.9)

(6.10)

(6.11)

where is the reaction rate of the substitutional carbon with the silicon self-intersti-

tial to form the interstitial-carbon and is the reaction rate of the interstitial-carbon

with the silicon vacancy to form a substitutional carbon. It is assumed that only the

is CIC →+

VCC is +→

CiVCsIs RR

t

C+−=

∂∂

CiVCsIiCii RRCD

t

C−+∇=

∂∂

CsII RIDt

I −∇=∂∂

CiVV RVDt

V −∇=∂∂

CsIR

CiVR

140

interstitial form of carbon is mobile and that the dominant mobile species is neutral.

The subsequent effect of local electric fields and extrinsic carrier concentrations on the

carbon diffusion is, therefore, neglected. This formulation was first proposed by

Scholz et al for carbon diffusion in silicon [14]. Ruecker et al applied the same model,

neglecting the contribution of germanium point defects to the carbon diffusion, and

relatively successfully fit carbon diffusion out of SiGe layers [16]. The reaction rates

are the sum of the forward and reverse reactions:

(6.12)

(6.13)

where and are the forward and reverse rate constants for the interstitial kick-

out mechanism and and are the forward and reverse rate constants for the

vacancy Frank-Turnbull mechanism. These rate constants may be estimated by

assuming a diffusion limited reaction, i.e. from the kinetic theory of non-interacting

particles:

(6.14)

(6.15)

where is the radius of the reaction cross section. Assuming no long-range interac-

tion, the cross section radius is on the order of an atomic radii. The reverse reactions

irIsfICsI CKICKR −=

srVifVCiV CKVCKR −=

fI rIK

fVK rVK

)(4 CsIrfI DDaK +⋅⋅⋅= π

)(4 CiVrfV DDaK +⋅⋅⋅= π

ra

141

are obtained assuming detailed balance in equilibrium and using measured equilibrium

quantities:

(6.16)

(6.17)

The ratio of is estimated from the ratio of measured diffusivities at equilib-

rium:

(6.18)

It can be shown that the total carbon out-diffusion from the SiGeC layer due to the KO

mechanism alone is dependent only on the transport product [13, 17, 18], i.e. it is

independent of for a set value of , but does depend on , which determines

the FT contribution to the carbon flux. is not well known because neither the

intrinsic vacancy concentration or diffusivity are well established, therefore, it is used

as the single adjustable parameter to fit the experiment. The critical parameters neces-

sary for this model are given below.

eq

i

sIr

eq

i

sfIrI C

CIDa

C

CIKK

⋅⋅⋅⋅≅

⋅⋅= )(4 π

eq

s

iCiVr

eq

s

ifVrV C

CVDDa

C

CVKK

⋅⋅+⋅⋅⋅=

⋅⋅= )(4 π

eq

i

s

C

C

1])[/1.3exp(9.1

])[/87.0exp(44.012

12

>>−⋅−⋅==

−

−

scmkT

scmkT

D

D

C

C

Cs

Ci

eq

i

s

ID I

fIK IDI rVK

rVK

142

Table 6.1. Model’s variables and their references

6.5.2 Modelling of carbon diffusion profiles

Because the structure from the previous section displays most of the critical observa-

tions, i.e. an interstitial sink in the carbon rich region while not depleting the intersti-

tial population between the SiGeC layers, it is used to illustrate the predictions of this

diffusion model. The time evolution of the diffused carbon profiles due to the com-

bined KO and FT mechanisms were simulated using as-grown carbon profiles

obtained from SIMS as the initial condition for the carbon concentration. The as-

grown carbon profiles was assumed entirely substitutional. The interstitial and

vacancy concentrations at the surfaces were set equal to the intrinsic concentration,

while the surface acted as a perfect sink for carbon, i.e. carbon diffusing to the surface

is removed from the sample (see section 6.2) for the nitrogen anneal case. For the oxi-

Variable Value Reference

I* 2.9x1024 exp(-3.18 eV/kT) [cm-3] [10]

DI 51 exp(-1.77 eV/kT) [cm2 s-1] [10]

V* 1.71x1024 exp(-2.0 eV/kT) [cm-3] [18]

DV 0.03 exp(-1.8 eV/kT) [cm2 s-1] [18]

Cs 4x1024 exp(-2.3 eV/kT) [cm-3] [6]

DCs 0.95 exp(-3.04 eV/kT) [cm2 s-1] [6]

DCi 0.44 exp(-0.87 eV/kT) [cm2 s-1] [6]

143

dation case, the same boundary conditions were used except that the surface interstitial

concentration was set to the enhanced value determined in the previous chapter, i.e.

12.7 times the intrinsic concentration I*. All other critical parameters are listed in

table 6.1. The simultaneous one-dimensional partial differential equations were

numerically solved by PROPHET [19]. The carbon profiles after annealing at 850ºC

for 30 minutes in nitrogen are shown compared to the simulation both overlaid on the

as-grown carbon profile (initial condition). In this case, when ~10-5 sec., enough

additional carbon outdiffusion is created to agree well with the experimentally

obtained carbon profiles, Fig 6.11.

rVK

144

Figure 6.11 Carbon concentration profiles from the double buried SiGeC layer sam-ple after annealing at 850°C in nitrogen ambient for 30 minutes overlaid on the profilecalculated using the combined kick-out and Frank-Turnbull mechanism.

The calculated interstitial carbon and silicon interstitial concentrations are dis-

played with the substitutional carbon profile for the annealing in nitrogen or oxygen

ambient for 30 minutes, (Fig 6.12 (a) and (b)) respectively. The interstitial concentra-

tion is depressed to less than a tenth of the intrinsic interstitial concentration (1.5x1010

cm-3) in the carbon rich region (Fig. 6.12 (a)) during nitrogen anneal. (Note: the self-

interstitial concentration is not well known. Estimates range over 4 orders of magni-

tude. The intrinsic self-interstitial concentration used in these simulations were those

1017

1018

1019

1020

1021

0 0.2 0.4 0.6 0.8 1

Car

bon

Con

cent

rati

on[c

m-3

]

Depth [µm]

850°C30 min

Simulation

SIMS

145

suggested by Bracht et al.) The silicon self-interstitial concentration remains near its

intrinsic concentration (when carbon is present) only as long as self-interstitials diffuse

in from the bulk and surface rapidly enough to supply the substitutional carbon reac-

tion with the silicon-self-interstitial. This is true when the substitutional carbon con-

centration is below the solid solubility concentration. In this case, however, the

substitutional carbon concentration is orders of magnitude greater than the typical

solid-solubility concentration. The high substitutional carbon concentration produces

an unusually high rate of self-interstitial consumption (eq’n. 6.12) that can not be sus-

tained by the transport capacity of the surrounding self-interstitials, which results in an

undersaturation of silicon self-interstitials in the high substitutional carbon region.

Figure 6.12 Simulated substitutional, interstitial carbon and interstitial silicon profilesafter 30 minutes of annealing at 850°C in (a) nitrogen or (b) oxygen ambient.

Con

cent

ratio

n[c

m-3

]

Con

cent

ratio

n[c

m-3

]

(a) (b)

850ºC, N230 min

850ºC, O230 min

10 7

10 9

10 11

10 13

10 15

10 17

10 19

10 21

0 0.2 0.4 0.6 0.8 1Depth [µm]

I

Cs

V

Ci10 7

10 9

10 11

10 13

10 15

10 17

10 19

10 21

0 0.2 0.4 0.6 0.8 1Depth [µm]

I

Cs

V

Ci

I*I*

146

In the previous chapter it was found that oxidation increases the surface con-

centration of silicon self-interstitials. Interstitials are injected into the bulk from the

surface due to the resulting self-interstitial gradient. The increased flow of self-inter-

stitials into the SiGeC layer does not, however, substantially increase the self-intersti-

tial concentration (above the self-interstitial concentration calculated for the nitrogen

anneal case) in the SiGeC layer 300 nm below the surface. The model predicts, there-

fore, that all injected interstitials react with the substitutional carbon at the SiGeC

layer agreeing with experiment. Furthermore, since the self-interstitials diffuse

quickly from the surface to the buried SiGeC layer (compared to the substitutional car-

bon) the self-interstitial concentration above the SiGeC layer reaches an approxi-

mately steady-state condition. The simulated interstitial profile, therefore, rapidly

becomes linear, which is consistent with experiment (chap. 5). The simulated “kick-

out” reaction, therefore, is sufficient to explain the experimentally observed “sink”

effect observed in chapter 5.

The interstitial concentration between the two SiGeC layers in both annealing

conditions rapidly increases to 3-4 times the intrinsic interstitial concentration

(1.5x1010 cm-3). The carbon diffusion simulation does not predict that the self-inter-

stitial concentration between the two SiGeC layers is depleted although all self-inter-

stitials between the carbon layers are well within the calculated intrinsic (no carbon)

self-interstitial diffusion length for this annealing condition (340 µm) [10]. The self-

147

interstitials have had time to diffuse to and react with the carbon in the surrounding

SiGeC layers. The simulation, therefore, shows that the self-interstitials consumed at

the carbon layer, from between the SiGeC layers, are somehow resupplied. Only one

reaction in the simulation produces self-interstitials, the reverse kick-out mechanism

(eq’n. 6.6). The simulation, therefore, shows that every self-interstitial that migrates

to and is consumed at the SiGeC layer is resupplied by the decay of an interstitial car-

bon.

Not only is the calculated self-interstitial concentration not depleted by the sur-

rounding SiGeC layers, but the simulation predicts an increase above the intrinsic con-

centration. More self-interstitials are released through the reverse KO reaction of

interstitial-carbon in the silicon region than self-interstitials are consumed by the

SiGeC layer. The increase in self-interstitial concentration above the intrinsic concen-

tration requires an additional source of self-interstitial generation. Since the model

produces self-interstitials only through the reverse KO reaction, the additional self-

interstitials are produced by another source of interstitial-carbon. The second source

of interstitial-carbon and increase of self-interstitial concentration above the intrinsic

concentration, in the calculation, is from the contribution of the Frank-Turnbull mech-

anism (eq’n. 6.7). As discussed in the previous sections (the experiment), the boron

diffusivity was depressed in the carbon rich regions but was found to be enhanced by

3-4 times between the SiGeC layers during annealing in nitrogen ambient. The

enhanced boron diffusivity between the SiGeC layers, which is an experimental mea-

148

sure of the enhanced self-interstitial concentration, agrees relatively well with the cal-

culation self-interstitial enhancement.

During oxidation the self-interstitial concentration between the two SiGeC lay-

ers is increased above that calculated during nitrogen anneal. Some of the extra inter-

stitial-carbon created by the extra injected self-interstitials migrate into the region

between the two SiGeC layers instead of leaving the sample. Most of these extra inter-

stitial-carbon become substitutional carbon releasing self-interstitials, which increases

the self-interstitial concentration above that in the nitrogen anneal case between the

two SiGeC layers. The calculation reproduces qualitatively the recycling effect pro-

posed in section 6.3, i.e. injected self-interstitials during oxidation are recycled in the

silicon bulk. However, little difference is observed between the measured (experi-

ment) boron diffusivities between the SiGeC layers during annealing in nitrogen or

oxygen ambient (Fig. 6.10). The calculation, therefore, overestimates the number of

self-interstitials that are recycled during oxidation for this structure.

In section 6.3 the recycling effect was suppressed in the oxidized sample with

the SiGeC layer buried only 60 nm below the silicon surface. To better understand

what determines the fraction of injected self-interstitials that are recycled a second

simulation of the carbon diffusion in this structure was performed. Using the same

model parameters as in the previous calculation, the self-interstitial, vacancy, substitu-

tional- and interstitial carbon concentration profiles after 30 minutes of oxidation at

850°C were simulated (6.13 (a)). From these profiles, the interstitial-carbon (solid

149

line), vacancy (dashed line) and silicon self-interstitial (dotted line) currents were cal-

culated from their respective diffusivities and simulated concentration gradients (Fig.

6.13 (b)).

Figure 6.13 (a) Simulated substitutional- , interstitial-carbon, self-interstitial andvacancy concentration profiles after 30 minutes oxidation at 850°C for structure B(SiGeC layer buried 60 nm below the surface, Fig 6.1 (b)) and (b) diffusion currentscalculated from simulated interstitial-carbon, silicon interstitial, and vacancy profiles.

First consider the interstitial-carbon and self-interstitial currents that decrease

from the SiGeC layer into the silicon bulk (Fig. 6.13 (b)). The decay of interstitial-car-

bon current with increasing depth into the bulk is a consequence of the interstitial-car-

bon decaying to substitutional carbon and self-interstitials. This decay feeds the

current of silicon interstitials back towards the carbon layer. The currents are almost

exactly balanced. The self-interstitial defect effectively runs a circuit flowing towards

the carbon layer as a free self-interstitial and then returning to the silicon bulk, trans-

-3x1010

-1.5x10 10

0

1.5x1010

3x1010

0 0.05 0.1 0.1Depth [µm]

(a) (b)

JI

JCi

JV

850°C, O230 min

Cur

rent

Den

sity

[ato

ms/

cm2

s]108

1010

1012

1014

1016

1018

1020

0 0.05 0.1 0.15

Con

cent

ratio

n[c

m-3

]

Depth [µm]

V

Cs

I

Ci

850°C, O230 min

150

ported as an out-going interstitial-carbon, until it decays and releases the self-intersti-

tial.

Now consider the currents in and out of the surface. The self-interstitial and

carbon-interstitial currents are nearly constant (indicating linear concentration gradi-

ents) and are considerably higher on the surface side of the SiGeC layer than on the

bulk side. The higher currents are created by the extra self-interstitials injected into

the silicon due to oxidation. After 30 minutes of oxidation, carbon has had time to dif-

fuse out the surface. As discussed earlier the surface is treated as a perfect sink for

carbon. The surface, therefore, creates an asymmetry in the interstitial-carbon profile,

which is steeper on the surface side (Fig. 6.12 (a)). The concentration gradient of

interstitial-carbon determines the interstitial-carbon current and the steeper concentra-

tion gradient on the surface side produces a greater current towards the surface than

into the bulk. The unequal currents lead to a disproportionate number of interstitial-

carbon that leave both the SiGeC layer and the sample compared to the interstitial-car-

bon that migrates into the bulk.

The calculation shows a nearly equal current of self-interstitials into the SiGeC

layer as the interstitial-carbon current out the surface at this oxidation time. This indi-

cates that most self-interstitials injected by oxidation are carried out of the sample as

interstitial-carbon. The difference in currents is ~ 1x109 cm-2 s-1 (self-interstitial cur-

rent is slightly greater than the interstitial-carbon current) indicating that less than 5%

of the injected interstitials are recycled in the bulk after reacting with the SiGeC layer

151

rather than are removed from the sample as interstitial-carbon flowing out the surface.

The model illustrates how a simple KO mechanism can lead to the disproportionate

loss of carbon from the SiGeC layer out the surface compared to the carbon that dif-

fuses into the bulk (i.e. almost every carbon “kicked-out” also leaves the sample as

observed in section 6.3 (Fig. 6.6 and 6.7)).

To summarize, the surface is considered a “sink” for interstitial-carbon. As the

SiGeC layer approaches the surface the interstitial-carbon gradient becomes much

steeper towards the surface than into the bulk. This causes a disproportionate amount

of interstitial-carbon to leave the SiGeC layer towards the surface compared to that

into the bulk. The recycling of self-interstitials depends on migration of interstitial-

carbon into the bulk followed by the reverse KO reaction to release a self-interstitial in

the bulk. As the carbon layer approaches the surface, the recycling effect is sup-

pressed because a greater fraction of the interstitial-carbon is driven out the sample

than goes into the bulk. The overestimated recycling calculated in the double SiGeC

layer structure, therefore, is due to an incorrect estimate of the interstitial-carbon gra-

dient into the bulk over the 30 minute oxidation time compared to that towards the sur-

face. The disagreement may simply be due to error in the choice of parameters within

this model like an underestimate of the interstitial-carbon diffusivity, which has only

been measured at low temperatures (below room temperature) [6].

To conclude, the diffusion model proposed, based on a simple kinetic theory

and using literature values for all critical parameters except one, illustrates how the

152

combined KO and FT mechanisms lead to very similar quantitative behavior as that

observed in experiment. The self-interstitial concentration is suppressed when the car-

bon concentration is raised above the solid-solubility, which would explain the

reduced boron diffusivity observed in SiGeC (chapter 3). The carbon oversaturated

silicon acts as a self-interstitial sink for injected self-interstitials (chapter 4 and 5) but

does not deplete the surrounding regions of self-interstitials because of the recycling

effect (chapter 6). Further work is required to test the validity of this model for differ-

ent annealing conditions and structures.

6.6 Summary

In conclusion, carbon out-diffusion from Si0.7865Ge0.21C0.0035 layers has

been examined after annealing in nitrogen or oxygen ambient at 850ºC. Carbon is

found to diffuse out the surface and the carbon diffusion from the SiGeC layer is

enhanced by oxidation. The extra number of substitutional carbon removed from the

SiGeC layer is explained by a “kick-out”-like reaction with the injected interstitials

during oxidation. The carbon out-diffusion from 25 nm thick SiGeC layer is found to

be the dominant mechanism of loss of substitutional carbon (as opposed to carbon

clusters or precipitation) in all annealing conditions examined for layers with silicon

caps less than or equal to 300 nm. Finally, carbon diffusion for concentrations above

solid-solubility has been modeled to illustrate how the combined KO and FT mecha-

nisms can lead to the same behavior as that observed experimentally. The simulations

153

support the conclusion that the self-interstitial reaction with the substitutional carbon

is “kick-out”-like and that the reaction mediates the substitutional carbon diffusion.

154

6.7 References

[1] J. W. Strane, H. J. Stein, S. R. Lee, S. T. Picraux, J. K. Watanabe, and J. W.Mayer, J. Appl. Phys., vol. 76, pp. 3656, 1994.[2] P. Warren, J. Mi, F. Overney, and M. Dutoit, J. of Crystal Growth, vol. 157, pp.414-419, 1995.[3] J. C. Sturm, P. V. Schwartz, E. J. Prinz, and H. Manoharan, J. Vac. Sci. Tech. B,vol. 9, pp. 2011, 1991.[4] D. De Salvador, M. Petrovich, M. Berti, F. Romanato, E. Napolitani, A. Drigo,J. Stangl, S. Zerlauth, M. Muehlberger, F. Schaeffler, G. Bauer, and P. C. Kelires,Phys. Rev. B, vol. 61, pp. 13005, 2000.[5] L. A. Ladd and J. P. Kalejs, presented at Oxygen, Carbon, Hydrogen, andNitrogen in Crystalline Silicon, 1986.[6] G. Davies and R. C. Newman, “Carbon in monocrystalline Silicon,” in Hand-book on Semiconductors, T. S. Moss, Ed., 1994, pp. 1558.[7] L. V. Kulik, D. A. Hits, M. W. Dashiell, and J. Kolodzey, Appl. Phys. Lett., vol.72, pp. 1972, 1998.[8] P. Werner, H.-J. Gossmann, D. C. Jacobson, and U. Goesele, Appl. Phys. Lett.,vol. 73, pp. 2465, 1998.[9] M. S. Carroll and J. C. Sturm, “Quantitiative measurement of the interstitialflux and surface super-saturation during oxidation of silicon,” presented at SiliconFront-End Processing-Physics and Technology of Dopant-Defect Interactions II, SanFransisco, 2000.[10] H. Bracht, N. A. Stolwijk, and H. Mehrer, Phys. Rev. B., vol. 52, pp. 16542,1995.[11] H.-J. Gossmann, C. S. Rafferty, H. S. Luftmann, F. C. Unterwald, T. Boone,and J. M. Poate, “Oxidation enhanced diffusion in Si B-doping superlattices and Siself-interstitial diffusivities,” Appl. Phys. Lett., vol. 63, pp. 639, 1993.[12] P. A. Stolk, H.-J. Gossmann, D. J. Eaglesham, D. C. Jacobson, C. S. Rafferty,G. H. Gilmer, M. Jaraiz, J. M. Poate, H. S. Luftmann, and T. E. Haynes, “Physicalmechanisms of transient enhanced dopant diffusion in ion-implanted silicon,” J. Appl.Phys., vol. 81, pp. 6031, 1997.[13] R. Scholz, U. Goesele, J. Y. Huh, and T. Y. Tan, Appl. Phys. Lett., vol. 72, pp.2, 1998.[14] R. F. Scholz, P. Werner, U. Goesele, and T. Y. Tan, Appl. Phys. Lett., vol. 74,pp. 392, 1999.[15] H. Ruecker, B. Heinemann, D. Bolze, D. Knoll, D. Krueger, R. Kurps, H. J.Osten, P. Schley, B. Tillack, and P. Zaumseil, presented at Technical Digest : Interna-tional Electron Device Meeting, San Fransisco, 1999.

155

[16] H. Ruecker and B. Heinemann, “Tailoring dopant diffusion for advancedSiGe:C heterojunction bipolar transistors,” Solid-State Elec., vol. 44, pp. 783, 2000.[17] A. Ural, P. B. Griffin, and J. D. Plummer, “Self-Diffusion in Silicon: Similar-ity between the Properties of Native Point Defects,” Phys. Rev. Lett., vol. 83, pp. 3454,1999.[18] H. Bracht, E. E. Haller, and R. Clark-Phelps, “Silicon Self-Diffusion in IsotopHeterostructures,” Phys. Rev. Lett., vol. 81, pp. 393, 1998.[19] M. R. Pinto, D. M. Boulin, C. S. Rafferty, R. K. Smith, W. M. Coughran, I. C.Kizilyalli, and M. J. Thoma, Tech. Digest IEDM, pp. 923, 1992.[20] Fahey, P. B. Griffin, and J. D. Plummer, Rev. Mod. Phys., vol. 61, pp. 289,1989.[21] H. Ruecker, B. Heinemann, W. Roepke, R. Kurps, D. Krueger, G. Lippert, andH. J. Osten, “Suppressed diffusion of boron and carbon in carbon-rich silicon,” Appl.Phys. Lett., vol. 73, pp. 1682, 1998.

156

Chapter 7

Low-Temperature Preparation of Oxygen- and Carbon-Free Silicon and Silicon-

Germanium Surfaces for Silicon and Silicon-Germanium Epitaxial Growth by

Rapid Thermal Chemical Vapor Deposition

7.1 Introduction

Small thermal budgets for silicon processing are increasingly demanded by ultra-large

scale integration (ULSI) technologies for various reasons such as the control of dopant

diffusion. The desire to integrate SiGe based heterojunction bipolar transistor (HBT)

technologies with silicon complementary metal-oxide-semiconductor (CMOS) tech-

nology has introduced a new demand for low thermal budget Si/SiGe epitaxy [1]. It is

now established that high quality silicon epitaxy can be grown by chemical vapor dep-

osition (CVD) at relatively low temperatures ranging between 500-800°C, despite

these temperatures being too low to desorb contaminants rapidly enough, such as oxy-

gen, from the surface of the silicon during the silicon epitaxy [2,3]. The low-temper-

ature CVD growth of silicon is achieved by relying on either low partial pressures of

contamination such as water vapor or hydrogen passivation to reduce the sticking

coefficient of the contamination. However, an initial clean substrate surface is neces-

sary for high-quality structures. Traditionally, CVD methods have relied on high tem-

perature in-situ cleaning steps such as hydrogen pre-bakes at 1000oC. Such

157

temperatures often are unacceptable if there are already dopant device profiles in the

substrate.

Several groups have reported oxygen- and carbon-free surfaces for subsequent

epitaxy prepared by in situ hydrogen pre-bakes between 760-850°C after an ex-situ

wet clean. However, there is special emphasis in these reports on ultra high vacuum

requirements and dry-pumped loading systems, and all of the reported cleaning proce-

dures depend on wet chemical preparation that are not compatible with the exposed

SiGe surfaces. In UHVCVD, no cleaning step at all beyond a wet ex-situ clean (e.g. an

HF dip) is required for high quality epitaxy at 760oC, but often residual carbon and

oxygen contamination still are found at the interface [4]. In this chapter we study the

dependence of surface quality on ex-situ wet cleaning and in-situ hydrogen baking

steps compatible with SiGe surfaces, without the need for UHV in the deposition

chamber. Secondary ion mass spectroscopy (SIMS) of buried interfaces and photolu-

minescence (PL) from thin buried SiGe layers are used to measure contamination. We

present a low pressure cleaning technique compatible with SiGe surfaces that reduces

oxygen and carbon contamination below the detection limits of SIMS for a rapid ther-

mal chemical vapor deposition system (RTCVD) using only conventional rotary vane

pumps.

This chapter is divided into four parts. First the standard procedure for epitax-

ial growth and interface characterization are described. Second, ex-situ wet cleaning

steps were examined. The third section focuses on contamination introduced by the

158

reactor and load lock. Finally the effect of in-situ bakes before growth is examined in

Section 4.

7.2. Standard Growth and Characterization Procedures

7.2.1 Cleaning and Growth Procedures

Standard growth procedures are now described. First the wafers are chemically

cleaned, beginning with the removal of the native oxide from p-type wafers (5-50

ohm-cm) using a ~5 min dilute HF dip (1:100 HF(49%):DI). The surface was then

chemically oxidized by immersing the wafer in H2SO4:H2O2(30%)1:1 at 70°C for 20

min. The oxide was then removed using an HF- based etch which leaves the surface

hydrogen terminated [5,6]. Unless otherwise noted, the HF(49%) to DI ratio for this

last step was 1:1000. After the last HF step, the wafer was not rinsed in DI water,

except when noted. The DI water resistivity was ~18 Megaohm-cm, the total organic

content (TOC) was <50 ppb, and the laboratory temperature was between 21-24°C

with a relative humidity below 50%. All chemicals were obtained from J. T. Baker

and were “CMOS electronic” grade. No precaution was taken to avoid dissolved oxy-

gen in the DI water [7].

Following the wet clean the wafer is placed on a quartz stand in the load-lock of

the growth reactor, which is evacuated to ~50 mtorr by a standard rotary-vane mechan-

ical pump. Three or four pump-purge cycles on the load-lock are performed before the

wafer is introduced to the growth chamber. The pump-purge cycle consists of filling

159

the load-lock to ~ 1 torr with dry nitrogen before evacuation, and one cycle takes an

average of approximately 5 minutes. The wafer is then transferred to the growth

chamber, which is kept between 1-10 torr of hydrogen.

The growth chamber is a cold wall system, using a bank of tungsten-halogen

lamps to radiatively heat the wafer through a quartz tube. Immediately following the

wafer transfer from the load-lock to the reactor, a flow of 1 slpm of hydrogen is passed

through the reactor while the reactor pressure was maintained at 1 torr. The hydrogen

is purified through a “Nanochem” [8] purifier, which is specified to reduce the impu-

rity concentrations in the hydrogen to less than 10 parts per billion (ppb). The growth

of silicon and silicon-germanium layers is done by rapid thermal chemical vapor dep-

osition (RTCVD) [3] using dichlorosilane and germane as source gases and hydrogen

as the carrier gas.

For our standard growth, a high temperature clean (to desorb oxygen) in hydro-

gen at 250 torr at 1000oC is then performed, followed by the growth of a high tem-

perature Si buffer layer at 1000oC. Low-temperature (e.g. 625oC for SiGe or 700oC

for Si) layers are then grown following the buffer. The 1000oC cleaning step and

buffer layer were frequently omitted or modified in the work described in this chapter,

since our goal was to develop cleaning steps which would not cause the diffusion of

any existing dopant profiles. Silicon and SiGe layers are typically grown at 6 torr in a

3 lpm hydrogen carrier with 26 sccm dichlorosilane and varying germane levels.

160

7.3 Characterization of Interfaces

All Si/SiGe interfaces were characterized using photoluminescence (PL) from the

pseudomorphically strained SiGe layers, which were immersed in a bath of liquid

nitrogen after growth. An argon ion laser tuned to 514 nm was used as the excitation

source, and a liquid-nitrogen-cooled Ge detector combined with a lock-in amplifier

was used to measure the emitted light. The pump power density was approximately

~50 W/cm2. Most of the minority carriers are generated in the substrate, and then dif-

fuse to the SiGe quantum well [9]. Therefore, the technique is best suited for struc-

tures without any barriers for carrier flow from the absorption region, a few microns

into the substrate, to the SiGe layer. A typical spectrum is shown in fig 7.1. Lumines-

cence intensity from the strained SiGe layer is extremely sensitive to the carrier life-

time in the SiGe layer [3,10]. Therefore any defects or contamination at the Si/SiGe

interfaces which lead to increased non-radiative recombination of excited carriers

reduce the overall luminescence intensity emitted from the Si1-xGex layer. The total

integrated SiGe luminescence was normalized to the total integrated luminescence

from the bulk silicon to account for any lifetime variation in the silicon substrates that

would affect the PL intensity from the SiGe layer.

161

Figure 7.1. Typical photoluminescent spectrum at 77 K of a “clean” Si0.8Ge0.2 200nm silicon layer capped with silicon. Sample (a) has an integrated contamination ofcarbon and oxygen below SIMS detection limits; while (b) has an integrated contami-

nation of 2x1012 cm-2 and 1x1014 cm-2 of carbon and oxygen respectively.

Some buried interfaces were also characterized using SIMS done at Evans

East, in East Windsor, NJ using a 3 keV Cs+ primary ion beam. Sputter rates were

between 5-15 Ångstroms/second, producing oxygen and carbon detection limits of

approximately 1018 cm-3 and 1017 cm-3 respectively for most samples. Sputter rates

were determined using profilometry leading to ~5% uncertainty in depth profiles and

chemical species concentrations were measured to within 15% error.

0

10

20

30

40

900 1000 1100Phot

olum

ines

cenc

eIn

tens

ity

[arb

.uni

ts]

Photon Energy [meV]

SiGe TO

SiGe NP

Si TO

(a)

(b)

162

Different cleaning procedures resulted in some interfaces that were contaminated

with carbon or oxygen. In these samples with higher oxygen or carbon levels at the Si/

SiGe interfaces, the PL intensity from the SiGe decreased, presumably due to recom-

bination sites caused by the contamination (fig. 7.1). Therefore, in this work the PL

intensity is often used as a probe of interfacial cleanliness since data can be obtained

more quickly than with SIMS. It also directly measures the electrical quality of the

SiGe.

7.4. Ex-situ wet cleaning

7.4.1 Introduction

In this section, the chemical wet treatment performed on the wafer before loading into

the reactor was studied and optimized. The treatment serves to remove the majority of

surface contaminants and to passivate the surface against further contamination. P-

type silicon (100) surfaces were prepared with a series of different wet cleaning meth-

ods. They were then loaded into the reactor, followed by growth of a thin ~200Å

Si0.8Ge0.2 layer without any in-situ cleaning step or buffer layers directly above the

wet-cleaned surface. A 450 Å silicon layer was subsequently grown above the

Si0.8Ge0.2 layer. Thus carriers in the Si0.8Ge0.2 layer are subject to effects of contamina-

tion on the original Si surface. In this set of experiments no in-situ high-temperature

cleaning process or high-temperature silicon buffer was grown, since our goal was to

study the effect of only the wet clean.

163

After transferring the wafer from the load lock to the growth chamber, the

hydrogen flow rate is increased to 3 slpm and the pressure is raised to 6 torr, followed

by heating the wafer to the growth temperature of 625°C over a period of 2 minutes.

When the growth temperature is reached, dichlorosilane is injected into the reactor

chamber (in addition to the already flowing hydrogen) followed by germane injection

approximately 5 seconds later, to grow the SiGe layer, at a growth rate of ~100 Å/min.

Following the growth of the SiGe layer, the Ge source was turned off and a Si capping

layer was grown after raising the temperature to 700oC for 15 minutes.

Germane is known to react with silicon dioxide to form the volatile species

GeOx. Oxide removal using germane at temperatures of 650-700°C has been reported

[11]. These reports found, however, that for sub-monolayer oxides, germanium

adsorbs preferentially to the bare silicon surface rather than forming the volatile ger-

manium-oxide [11]. The sequence of dichlorosilane followed by germane was chosen

to allow the SiGe layer to grow as soon as the germane is injected. Because the SiGe

layer grows quickly on the silicon surface with no observable long incubation time,

and the germane molecule prefers the open silicon surface site over that of the oxide

site, it is concluded that the germane-induced oxide desorption has a negligible effect

on the total oxygen concentrations buried at the interface between the silicon surface

and the SiGe layer.

Following the initial HF dip and chemical oxidation as described earlier, the

concentration (and thus pH) for the final HF step before loading was varied. The

164

wafer was dipped in a final oxide etch with a pH of 1-7.8 for a minimum of 20 min-

utes. The HF to DI ratio for this final oxide stripping step was varied from 1 to a 1000

parts DI to one part HF (49%) (pH between 1 and 3) to determine the effect of the HF

concentration. Less acidic solutions were also examined, using HF buffered with

NH4F (pH ~4) or slightly basic 40% NH4F (pH~7.8) instead of HF and water.

Although the wafer surface becomes hydrophobic at shorter times (~10 minutes in

1:1000 HF(49%) in DI), the minimum 20 minute etch time was chosen to insure suffi-

cient time for the oxide removal reaction to go to completion. PL from Si1-xGex layers

grown above surfaces that were rinsed in DI after an HF dip (1:1000 HF(49%) in DI)

were completely quenched. Presumably, dissociated OH- ions oxidize the silicon sur-

face [12,13] resulting in a poor Si/SiGe interface, which quenches the PL from the Si1-

xGex layer. Therefore, after the dilute HF dip step any residual droplets on the mostly

dry surface were blown off with nitrogen, but not rinsed. However, no SIMS or fur-

ther work was done to confirm this hypothesis. The hydrogen passivated surface was

then exposed to laboratory atmosphere for 1-20 minutes before loading into the load-

lock. High quality interfaces were achieved even after 15-20 minutes of exposure to

air, indicated by intense PL from Si1-xGex layers grown above the exposed surface.

A monotonic increase in the relative PL intensity from the Si0.8Ge0.2 layer was

found as the HF concentration was decreased (fig. 7.2), indicating less contamination.

However solutions more basic than the most dilute HF dip led to a decrease in lumi-

165

nescence intensity from the Si0.8Ge0.2 layer (pH>3). This optimum treatment of

H2SO4/H2O2 (70oC, 20 min.) followed by a 1:1000 HF/DI dip (after removing the

native oxide using a 1:100 HF/DI dip) was also applied directly to a Si0.8Ge0.2 surface.

The Si0.8Ge0.2 surface was grown at 625oC on top of a 1000oC Si buffer followed by

removing the exposed Si0.8Ge0.2 surface from the reactor and leaving it in the labora-

tory atmosphere for 3 days. The Si0.8Ge0.2 surface was then treated with the same wet

clean as that used for the pure silicon surfaces followed by reinserting the Si0.8Ge0.2

surface into the reactor for the 700oC growth of a Si cap without any hydrogen baking.

The PL intensity from the resulting structure was more intense, 7.4, than the case

where the pure silicon surface was cleaned followed by a silicon capped Si0.8Ge0.2

layer, 3.8 (fig. 7.2). Thus we conclude the combination of H2SO4/H2O2 and 1:1000

HF:DI dip is also effective in preparing clean SiGe surfaces.

166

Figure 7.2. Relative photoluminescence SiGe/Si (PL) intensity at 77K from com-mensurately strained Si0.8Ge0.2 alloys followed by silicon caps grown directly abovewet-chemically-treated silicon surfaces vs. the pH of the solution used to etch the wetchemical oxide on silicon. The ratios of either HF (49%) to DI water, or NH4F (40%)to HF (49%) are indicated above their respective points. In one case the HF dip was

also done instead on top of SiGe, followed by a 700oC Si cap.

SIMS was available for silicon surfaces, which were not subjected to the stan-

dard wet chemical clean, but rather were prepared by silicon growth in the RTCVD

reactor. The wafers were then removed from the reactor and immersed for various

times in either 1:10 or 1:100 HF(49%):DI (pH’s = 1.4 or 1.9, respectively). The sam-

ples were then returned to the reactor through the standard loading procedure. The

surfaces were then buried under an epitaxial SiGe layer and Si cap as described earlier

0.1

1

10

1 2 3 4 5 6 7 8

Rel

ativ

eP

LIn

tens

ity

I SiG

e/I

Si

pH of Etch Solution

1:1000

1:10

1:100

1:1 HF:DI NH4F:HF

1:0

10:1

167

in this section. The integrated carbon and oxygen levels are ~5 times higher for a pH

of 1.4 than 1.9, and fluorine is only found for pH of 1.9 (fig. 7.3).

Figure 7.3. The carbon, oxygen, and fluorine concentrations detected by SIMS at Si/

SiGe interfaces from SiGe layers grown directly at 625oC on Si surfaces treated withHF:DI solution of pH 1.4 and 1.9 to remove the wet chemical oxide, without a hydro-gen pre-bake.

This decrease in contamination as pH is raised (to at least to pH of 3) may be

related to the fact that (111) micro-faceting of the (100) silicon surface is known to

increase with increasing pH [14], due to anisotropic etching by OH- ions [15,16]. The

resulting monohydride-terminated (111) surface has an oxygen sticking coefficient

~100 times lower than the dihydride terminated (100) hydrogen-passivated surface

1010

1011

1012

1013

1014

1.9 1.4

Con

cent

ratio

n[c

m-2

]

pH of Etch Solution

O

OC

C

F

168

[17,18,19]. Therefore less contamination and an increase in SiGe PL would be

expected with increasing pH and increased microfacetting.

The increased fluorine contamination with increasing HF concentration has previ-

ously been observed [20,21,22]. The observed increase in carbon contamination at

low pH could also be due to organic content in the process chemicals, which would

increase with increased process chemical concentrations [20,21]. The reason for the

decrease in SiGe PL at pH greater than 3 is not known. Silicon substrates cleaned

with NH4F have been reported to have an increased density of crystallographic defects

in the epitaxial silicon grown above the prepared surface [23], which could be respon-

sible for the reduced PL. However, the epitaxial films were not examined for crystal-

lographic defects, therefore, different surface termination or impurities from the NH4F

solution can not be ruled out as causes for the PL quenching.

7.5. Contamination from Reactor, Load-Lock, and Laboratory Environment

7.5.1 Experiment

In our system, wafers are introduced to the growth chamber through a load lock, which

is pumped by a rotary vane pump using hydrocarbon-based oil. The load-lock was

therefore examined as a source of carbon contamination. Clean pseudomorphically

strained Si0.8Ge0.2 surfaces were prepared by growing a silicon epitaxial buffer layer

followed by a thin 200Å Si0.8Ge0.2 layer. Immediately after growth, the chamber was

purged with a continuous flow of 3 slpm of hydrogen at 6 torr. The wafer is allowed to

169

cool for 15 minutes before being moved to the load-lock where parts or all of the load-

ing procedure were simulated. After transferring back to the growth chamber, the

wafer was then heated to 700oC (ramp rate 50°C/sec) in 3 lpm H2, and after ~15 sec at

700oC dichlorosilane was switched on to grow an epitaxial silicon cap (45 nm). No

high-temperature cleaning steps in excess of 700oC were used. The upper SiGe/Si

interface was then examined by SIMS and PL to look for contamination introduced by

the load-lock.

Four experiments on SiGe surfaces were done before returning the wafer to the

deposition chamber for a capping silicon layer:

Trial A: leave wafer in reactor;

Trial B: transfer wafer to load-lock + simulate load-lock pump-down;

Trial C: trial B + vent load-lock to atmosphere + load-lock pump-down;

Trial D: trial C, except the wafer is moved to the fume hood;

To first determine whether the growth chamber itself contributes any contamina-

tion to the surface (apart from the load-lock), the SiGe layer wafer A was simply left

in the growth chamber without transfer to the load lock . The wafer was allowed to

cool in 3 slpm of H2 at 6 torr for 15 minutes. This time exceeds the time required to

transfer the wafer to the load-lock and therefore simulates a maximum contribution of

the growth chamber’s contamination to the wafer surface. The wafer was then

reheated to 700oC and a silicon cap was grown as described above. No oxygen or car-

170

bon is detected by SIMS at the interrupted SiGe/Si interface (figs. 7.4, 7.6). This layer

exhibited intense photoluminescence also indicating negligible contamination at the

interface due to the 15 minutes spent in the reactor chamber between layers.

Figure 7.4. Oxygen, carbon and germanium SIMS profiles of a sample in which acommensurately strained Si0.8Ge0.2 surface prepared in-situ by rapid thermal chemicalvapor deposition, then allowed to cool for 15 minutes with 3 slpm of hydrogen flowingat 6 torr before burying the test surface with silicon epitaxy using dichlorosilane at700ºC (Wafer “A”). The hydrogen bake was done at the silicon depth of ~700Å.

Wafer B was transferred to the load-lock after the SiGe growth. A purge cycle,

consisting of pressurizing the load-lock to 1-10 torr with dry nitrogen followed by

evacuation with a rotary vane pump, was repeated 6 times to simulate the load-lock

transfer without exposure to atmosphere. SIMS of the buried interface detected barely

1017

1018

1019

1020

1021

1022

1023

600 700 800 900 1000

CarbonOxygenGermanium

Con

cent

rati

on[c

m-3

]

Depth [Ångstroms]

H2 BakeLocation

171

observable C and O peaks (fig. 7.5 at depth of 0.13 µm) with integrated densities in

both cases <1012 cm-2. To simulate the entire loading process from removal of the

substrate from the ex-situ dilute HF dip to loading and purging of the load-lock, two

more cases were examined: (Wafer C) a strained SiGe layer was transferred to the

load-lock, given a nitrogen purge cycle, left in the load-lock for 5 minutes with the

door of the load-lock left slightly open to the laboratory atmosphere, and then sent

through a second purge before being returned to the growth chambers. After SiGe

growth, Wafer D was taken out of the load-lock and moved to a chemical hood for 10

minutes, after which the wafer was returned to the reactor through the standard loading

and purge cycles without any HF dip or wet processing.

172

Figure 7.5. Oxygen, carbon and germanium profiles at SiGe/Si interfaces when thesample was transferred to the load lock (B) without venting to air, and (C) with vent-ing to air, before silicon cap regrowth at 700ºC.

SIMS of Wafer B to (nitrogen only atmosphere in the load lock) and of Wafer

C (door of load lock slightly opened) showed only a small increase in contamination

(figures 7.5, 7.6). However when the wafer was removed from the load-lock and

exposed to atmospheric conditions of the laboratory (Wafer D), the carbon and espe-

cially the oxygen contamination increased significantly (fig. 7.6). Finally, to compare

the susceptibility of SiGe surfaces to Si surfaces for contamination, in one case (Wafer

E) a virgin Si wafer was left in the fume hood for 10 minutes after a 100:1 DI:HF dip

before loading and the growth of a silicon cap at 700oC. Oxygen and carbon concen-

1017

1018

1019

1020

1021

1022

0.05 0.1 0.15 0.2 0.25

OxygenCarbonGermanium

Con

cent

rati

on[c

m-3

]

Depth [µm]

Vent Load-Lock (C)

Transfer toLoad-Lock (B)

173

trations for this wafer are 5-10 times less than those of Wafer D (fig. 7.6), in which a

presumably clean and H-terminated SiGe surface from the reactor was exposed simi-

larly to air for ten minutes.

Figure 7.6. Integrated oxygen and carbon levels observed at SiGe/Si interfaces bySIMS for wafers (A-D) subjected to different conditions after SiGe growth before Sicap growth at 700ºC without high temperature cleaning. In case E, the test surface isnot prepared in-situ as in cases A-D, a silicon substrate surface after 1:100 HF:DI dipis transferred to the load-lock (total time in fume-hood ~10-15 minutes) before silicongrowth at 700ºC. Note: solid and dashed lines indicates SIMS detection limits foroxygen and carbon respectively (SIMS background multiplied by typical contamina-tion peak widths).

1010

1011

1012

1013

1014

1015

A B C D E

OxygenCarbon

Inte

grat

edIn

terf

ace

Con

tam

inat

ion

[cm

-2]

O BackgroundC Background

Leave inReactor

(A)

Load-Lock(B)

VentLoad-Lock

(C)

FumeHood(D)

Wet CleanHF:DI (1:100)

(E)

174

7.5.2 Discussion

Several conclusions can be drawn from this series of experiments. First, the

reactor itself is a relatively clean environment for short times for cold hydrogen passi-

vated surfaces. Most significantly, the load-lock itself introduces only minimal con-

tamination. Most contamination is caused by the exposure of the wafer surface to the

laboratory environment. Finally, the hydrogen-terminated surface of the in-situ pre-

pared SiGe surface is much more susceptible to oxygen and carbon absorption than a

hydrogen- terminated Si surface. The difference in reactivity of the in-situ prepared

SiGe surface compared to the surface prepared in a 1:100 dilute HF dip may be due to

differences in the stability of the hydride termination of the two cases, which can result

from different reconstructions of the surface [24], or because the H-Ge bond is weaker

than the H-Si bond [25].

7.6. In-situ Wafer Cleaning

7.6.1 200-400°C pre-bakes

The purpose of a very low-temperature (e.g. 200-400oC) bake is to desorb physisorbed

chemical contamination from the surface before the chemical contamination can dis-

sociate and chemisorb to the wafer surface, which occurs first at higher temperatures

[26]. Any chemisorbed carbon that is consequently annealed at high temperature can

form stable SiC precipitates on the surface [27] or diffuse into the bulk [27], leading to

undesirable defect formation.

175

To test the effectiveness of low-temperature pre-bakes on the interface quality,

(100) wafers were subjected to the standard wet clean and loading procedure, followed

by a low-temperature pre-bake before epitaxy. No high-temperature bake was used.

The wafers were baked at ~300°C at 6 torr under a hydrogen flow of 3 lpm. A 200Å

Si0.8Ge0.2 layer was grown at 625oC immediately after the hydrogen bake, followed

by a 450 Å silicon capping layer to reduce surface recombination effects on photolu-

minescence intensity [28]. The time dependence of the interface quality on the hydro-

gen pre-bake was examined using photoluminescence intensity (fig. 7.7). Because the

PL intensity decreased with extended low temperature pre-baking it was concluded

that the surface quality is degraded, not enhanced by a low-temperature pre-bake in

our system.

176

Figure 7.7. Relative SiGe/Si photoluminescence intensities from SiGe/Si layersgrown directly on silicon (100) substrates prepared using 1:1000 HF(49%):DI andthen baked in 6 torr of H2 at 300ºC for different times before SiGe growth.

7.6.2 700-800°C Bakes in Hydrogen

The complete removal of oxygen and carbon from silicon surfaces before epitaxy by

baking in hydrogen atmospheres at higher temperatures (750-850°C) has been previ-

ously reported for UHV-CVD [4,29]. Their success has been attributed to low oxygen

and water-vapor partial pressures, below the critical levels required for clean silicon

surfaces in a vacuum at a given temperature [30,31]. Ref’s. [4] and [29] stressed the

importance of both the UHV system and the hydrocarbon-free “dry” load-lock system.

Later, Wolansky et. al. demonstrated that oxygen and carbon removal could be

0

2

4

0 30 60 90

Rel

ativ

eSi

Ge/

SiPL

Inte

nsit

y

Bake Time [min]

300°C6 torr H

2

177

obtained at higher hydrogen pressures for nearly equivalent thermal budgets [1],

showing that UHV is not a necessary condition for the cleaning of the silicon surface

in hydrogen at or under 800oC, and that the cleaning temperature could be reduced to

as low as 760°C. However, some carbon contamination was seen at the cleaned inter-

face by SIMS in that report. In our work, hydrogen bakes were examined for different

pressures and temperatures ranging from 0.5-250 torr and 700 to 800°C.

7.6.2.1 700°C Bakes in Hydrogen

The growth of silicon epitaxial layers at 700oC was interrupted, after which the sub-

strates were then allowed to cool to less than 200°C in 3 slpm hydrogen at 6 torr in the

reactor. In Sec. 4.1, we showed this introduced no contamination to a SiGe surface. A

silicon surface would presumably also remain clean due to the higher stability of

hydrogen on its surface (Sec. 4.1). After cooling, for approximately 15 minutes, the

reactor pressure was set between 0.25-250 torr, with a hydrogen flow of 3-4 lpm,

immediately followed by reheating the wafer to 700°C. The temperature, flow, and

pressure were maintained constant for between 2-15 minutes, with the exception of the

hydrogen bake at 0.25 torr in which the hydrogen pressure was held constant by turn-

ing off the hydrogen flow and sealing the chamber. After the hydrogen bake the pres-

sure was then brought to 6 torr with a flow of 3 slpm of hydrogen. Dichlorosilane was

then added to the hydrogen stream to grow epitaxial silicon and bury the test interface.

The oxygen and carbon absorption rates during the bake were determined by dividing

178

the integrated carbon and oxygen concentrations found at the interface (determined by

SIMS), by their respective annealing times. At lower pressure, for example 0.6 torr,

the absorption rates for oxygen and carbon were 1.6x1014 cm-2 min-1 and 1.4x1012

cm-2 min-1 respectively (Table 1). However, for pressures of 6 and 250 torr of hydro-

gen both oxygen and carbon concentrations were below SIMS detection limits. For

the 6 torr case the adsorption rates were less than 7x1010 cm-2 min-1 and 7x109 cm-2

min-1, and for 250 torr the detection limits were anomalously high leading to a higher

maximum bound on the adsorption rates, which were 5x1012 cm-2 min-1 and 5x1011

cm-2 min-1 for oxygen and carbon respectively

179

.

While the bakes at pressures at or over 6 torr did not contaminate the surface, in

our lab they have also not been effective at removing previously existing contamina-

tion, presumably because the temperature is too low for rapid enough desorption. A

previously prepared SiGe surface exposed to atmosphere for over three days was pre-

pared by wet cleaning, using the optimal 1:1000 HF(49%):DI ratio. The wafers were

loaded into the RTCVD for the growth of a silicon cap at 700oC. The quality of the

Table 1: Integrated oxygen and carbon interface levels measured by SIMS andinferred adsorption rates for Si surfaces baked in hydrogen at pressures from 0.25 to250 torr for various times. All hydrogen pressures were maintained with a steady flow

rate of 3-4 lpm of hydrogen and baked at 700oC except for 0.25 torr, which was bakedwith no hydrogen flowing. Less than or equals indicates no O or C detected within theresolution of SIMS and the given rate is calculated from the limit of SIMS backgroundfor the sample.

Pressure

[torr]

Time

[min]

Oxygen

[cm-2]

Carbon

[cm-2]

Oxygen

Adsorption

Rate [cm-2

min-1]

Carbon

Adsorption

Rate [cm-2

min-1]

0.25 2 8.7 x 1014 2.8 x 1012 4.4 x 1014 1.4x1012

0.6 15 1.6 x 1015 4.9 x 1012 1.0 x 1014 3.2x1011

6 15

250 2

1x1012≤ 1x1011≤ 7x109≤ 7x109≤

1x1013≤ 1x1012≤ 7x109≤ 5x1011≤

180

SiGe layer and its interfaces were probed by PL. Performing a 10 min bake at 700oC

at 6 torr in 3 lpm H2 before the Si epitaxy showed no improvement in the SiGe/Si PL

intensity compared to that of a wafer grown without such a step. Also, there was no

observed PL dependence on when the dichlorosilane was introduced (i.e. either before

or after the wafer was heated to 700°C). Further work is needed to evaluate longer

700°C hydrogen bake times and pressures for any potential cleaning benefits.

7.6.2.2 800°C Bake in Hydrogen

Since 700oC bakes were not effective, 800oC hydrogen bakes were tested to remove

oxygen and carbon from the silicon surface after wet cleaning. Silicon surfaces were

cleaned ex-situ before introduction to the reactor, as described before using the opti-

mal 1000:1 DI/HF final dip (Sec. 3). After loading, the hydrogen flow is set to 3-4

lpm and the hydrogen pressure in the reactor is brought to between 0.5 and 250 torr.

The wafer was then heated and held at 800°C for 1 minute, after which the temperature

is reduced to 625°C while stabilizing the pressure to 6 torr (at 3 lpm H2). After both

pressure and temperature are stable (requiring ~ 1 minute for all pressures except 250

torr which required ~ 5 minutes), DCS followed shortly after by germane were

injected into the reactor to grow a Si0.8Ge0.2 (20nm) layer, followed by a Si capping

layer at 700°C. PL results, and integrated oxygen and carbon concentrations (by

181

SIMS) at the lower Si/SiGe interface for different pressures and for different times at

800°C are shown in figures 7.8, 7.9, and 7.10, respectively.

Figure 7.8. Relative SiGe/Si photoluminescence intensities from SiGe/Si layers(20nm/45nm) grown on silicon (100) substrates prepared using 1:1000 HF(49%):DIand then baked for 1 minute at 800ºC at different hydrogen pressures.

If the hydrogen bake is omitted (wafer E, fig. 7.6), the integrated O and C lev-

els are 2.3x1013 cm-2 and 4x1012 cm-2. Performing a hydrogen bake for 1 min at 0.5

torr slightly reduces the integrated oxygen level to 6.4x1012 cm-2 and increased the

carbon level to 6.6x1013 cm-2. The 1 minute 250 torr bake dramatically increases the

oxygen level to 9x1013 cm-2 and drops the carbon level to 2x1012 cm-2. The 1- and

0.1

1

10

100

0.1 1 10 100

Rel

ativ

eS

iGe/

SiP

LIn

tens

ity

H2

Pressure [torr]

800ºC1 min

182

10-torr bakes both reduced the oxygen to ~1012 cm-2, and the carbon levels were little

changed from the no-bake levels (~10 13 cm-2). The SiGe/Si relative PL intensity

dropped 10-100 times for 0.5 or 250-torr cleans, corresponding to the 10-100x

increase in oxygen or carbon levels, therefore, the SiGe/Si PL ratio was consistent

with SIMS in showing that the 1-10 torr cleans, which also had the lowest oxygen and

carbon contamination, had the highest SiGe/Si PL ratios. The 1-10 torr bakes

increased the SiGe/Si PL ratios from 1-4 without a bake to 9-12; this is in comparison

to a ratio of 10-20, which is commonly observed for Si/SiGe/Si structures grown under

our best conditions (grown without interruption, to minimize contamination, after a

1000oC hydrogen clean and silicon 1000o buffer layer on the substrates).

183

Figure 7.9. Integrated carbon and oxygen levels at SiGe/Si interfaces formed bygrowing SiGe on silicon (100) substrates prepared using HF dip and then baked for 1minute at 800ºC at different hydrogen pressures.

A longer hydrogen bake of 2 minutes at 800°C and 10 torr results in no detect-

able oxygen or carbon contamination above the SIMS background (fig.7.10), although

the SIMS oxygen background was considerably higher in this case. The interrupted

and uninterrupted growth are, therefore, indistinguishable with respect to PL and

SIMS measurements. This cleaning technique (800oC, 2 min, 10 torr H2) was also

used in a Si/Si interface after ex-situ wet cleaning, followed by 700oC Si growth where

a phosphorus layer marked the location of the interrupted interface. Again, absolutely

1011

1012

1013

1014

0.1 1 10 100 1000

CarbonOxygen

Con

cent

rati

on[c

m-2

]

H2

Pressure [torr]

C Background800°C1 min

184

no carbon or oxygen was detected by SIMS at this interface, even with the integrated

concentration detection limit due to SIMS background as low as 1012 cm-2 for oxygen

and 4x1011 cm-2 for carbon (fig. 7.11).

Figure 7.10. Integrated carbon and oxygen levels at Si/SiGe interfaces formed SiGegrowth on silicon (100) substrates prepared using an HF dip to remove the wet chemi-

cal oxide and baked at 800oC in 10 torr hydrogen for various times. For a two minutebake, the oxygen level was below that resolvable by SIMS.

7.6.3. Discussion

The net oxygen or carbon adsorption or desorption to or from the silicon surface at dif-

ferent temperatures and hydrogen pressures depends on the flux to the surface, the

sticking fraction of the flux that sticks to the surface, and the desorption rate. Hydro-

gen passivation of the surface, which increases at increasing hydrogen pressure, can

1012

1013

1014

0 1 2

CarbonOxygenCarbon wet clean (1% HF) onlyOxygen wet clean (1% HF) only

Inte

grat

edC

once

ntra

tion

[cm

-2]

Bake Time [min]

C BackgroundO Background

800°C10 torr

185

greatly reduce the sticking coefficient of oxygen [32]. However, a simple hydrogen

passivation model fails to explain why during an 800°C bake, the adsorbed oxygen

increases as the hydrogen pressure is increased from 6 torr to 250 torr (fig. 7.9). The

increase in oxygen contamination at higher hydrogen pressure could be the result of

oxygen or water vapor impurities in the hydrogen gas. The two sources of oxygen/

water contamination are then the background in the reactor and the carrier gas itself.

Because the hydrogen entering the reactor is purified to a level 10 ppb, the hydrogen

gas can only become the dominant source of oxygen at high hydrogen pressures.

186

Figure 7.11. Oxygen, carbon and phosphorus SIMS profiles of a sample in which thephosphorus doped silicon marks the interrupt location, where the silicon surface iscleaned using the experimentally determined optimal ex-situ and in-situ conditions

(i.e. using a 1:1000 HF(49%):DI to strip the final chemical oxide followed by a 800oC,10 torr hydrogen bake for 2 minutes). The silicon cleaned interface is located at adepth of approximately 5050 Å.

Figure 7.12 schematically shows the two contributions to oxygen adsorption of

hydrogen surface coverage and oxygen partial pressure, and their dependence on

hydrogen pressure. The author stresses that this description is only a qualitative

description, as details of the hydrogen surface coverage, and total oxygen background,

are not exactly known. As the hydrogen pressure increases, the hydrogen coverage of

the surface increases, which greatly reduces the number of open sites for O absorption

1017

1018

1019

4800 5200 5600

PhosphorusOygenCarbon

Con

cent

rati

on[c

m-3

]

Depth [Ångstroms]

InterruptLocation

187

and thus the sticking coefficient [32]. This explains the relatively high contamination

levels resulting from bakes at pressures below 6 torr at both 700 and 800oC. When

the hydrogen pressure is too high, however, then the oxygen partial pressure, due to

impurities in the hydrogen, may become so great that it produces an oxygen flux that

can not be compensated for by the additional hydrogen coverage. At 700oC the hydro-

gen coverage at high pressure is still sufficient to keep the increased oxygen contami-

nation in the gas below detection limits (SIMS detection limits were high in this case,

Table 1). However, because silicon epitaxy in this reactor at higher hydrogen pressure

(220 torr, 700oC) has shown indications of unusually high oxygen concentrations [10]

and the oxygen SIMS detection limits were high (because of interference with surface

contamination), it is still likely that the surface may be adsorbing oxygen at higher

hydrogen pressures at 700oC. At 800oC the higher open site density vs. that at 700oC

allows the increased gas contamination level to cause significant surface contamina-

tion, so that high pressure bakes contaminate the surface (fig. 7.9).

188

Figure 7.12. Schematic diagram of important mechanisms that determine the amountof oxygen sticking to the silicon surface during hydrogen baking. The decreasing frac-tion of open sites that are available for oxygen to stick (corresponding to increasedhydrogen coverage) with increasing hydrogen pressure at 700ºC and 800ºC are refer-enced to the left axis. Oxygen partial pressures due to the oxygen background in thereactor and from impurities in the hydrogen gas are referenced to the right axis.

At 800oC, the cleanest surface is achieved between those two extremes, in the

1-10 torr range. The inability to further clean surfaces at 700oC, while still maintain-

ing low contamination at high pressures (oxygen and carbon below SIMS detection

limits), can be attributed to an oxygen desorption rate near the rate of oxygen adsorp-

tion. If the accumulative rate is indeed desorptive at this temperature it is still too slow

to make an observable effect for the bake times considered in this study.

0.0001

0.001

0.01

0.1

1

0.1 1 10 10010-9

10-8

10-7

10-6

10-5

Frac

tion

ofO

pen

Site

s[a

rb.u

nits

]

Reactor Background

H2

Coverage

(T=800°C)

H2

Coverage

(T=700°C)

contribution fromimpure H

2

H2 Pressure

OP

artialPressure

Ope

nSi

tes

189

The surface concentration of carbon is also observed to have a dependence on

hydrogen pressure. At the lowest hydrogen pressure of 0.5 torr, the detected carbon

signal rises significantly, 6.6x1013 cm-2, but at higher hydrogen pressures, 250 torr, the

carbon signal drops to below the SIMS detection limits, 2x1012 cm-2 after 1 minute at

800oC (fig. 7.9). The source of carbon depositing on the surface during baking at low

hydrogen pressure, 0.5 torr, is likely from reactor background contamination. Presum-

ably increasing the hydrogen coverage of the surface, by increasing the hydrogen pres-

sure, will reduce the overall adsorption of carbon from the chamber atmosphere.

Indeed, less carbon is observed for higher hydrogen pressure bakes, however, it is also

observed that carbon is removed from the silicon surface for high hydrogen pressures,

10 torr and 250 torr, at 800oC. Two proposed mechanisms of carbon removal from the

silicon surface during hydrogen baking at temperatures around 800oC are either des-

orption of carbon as hydrocarbons or methylsilanes [4, 29], or diffusion of carbon

from the surface into the silicon bulk [27]. However, further analysis of these two pos-

sible mechanisms goes beyond the scope of this work due to limits of SIMS resolution

(i.e. SIMS broadening and detection limits) and an incomplete knowledge of how

much carbon is desorbed into the hydrogen atmosphere.

190

7.7. Summary

Photoluminescence (PL) from commensurately strained SiGe layers grown directly on

silicon substrates and secondary ion mass spectroscopy (SIMS) of buried Si/SiGe

interfaces are used to evaluate different low-temperature cleaning methods of substrate

surfaces for silicon and SiGe epitaxy in a non-ultra-high vacuum system. Both the

sources of contamination as well as effective cleaning methods were investigated. The

dominant source of contamination came from the wafer being outside the reactor, not

in the load lock or deposition chamber itself. The optimum surface preparation

depends on the ratios of HF, NH4F and de-ionized water of solutions that were used to

remove the wet chemical oxide on the substrate surface. In-situ bakes between 300ºC

and 800ºC in 0.25-250 torr of hydrogen were examined after the ex-situ clean using

PL and SIMS measurements. 700oC bakes do not add contamination at sufficiently

high hydrogen pressure, but are also ineffective at removing existing oxygen and car-

bon contamination. 800oC bakes between 1-10 torr can effectively remove contami-

nation and give interfaces which are indistinguishable by SIMS or photoluminescence

from those grown without interruption.

The pressure dependence of the interface cleaning at 700oC and 800oC may be

understood by considering the effect of the hydrogen pressure on the reactivity of the

surface. Because negligible dopant diffusion occurs for short times at 800oC, this

demonstrated ability to grow pristine interfaces without exceeding 800oC after remov-

191

ing the wafer from the reactor will enable new strategies for device integration and

fabrication, which in one case has already been successfully demonstrated in a vertical

MOSFETs structure [34].

192

7.8 References

[1] D. Wolansky, B. Tillack, K. Blum, K. D. Bolze, K. D. Glowatzki, K. Köpke, D.

Krüger, R. Kurps, G. Ritter, P. Schley, in Proceedings of the 8th International Sympo-sium on Silicon Material Science and Technology. Silicon Material Science and Tech-nology, H. R. Huff, H. Tsuya, U. Gosele, Editors, p. 812-821. The ElectrochemicalSociety Proceedings. Pennington. NJ (1998)[2] D. W. Greve, Mat Sci. & Eng., B18, 22-51 (1993)[3] J. C. Sturm, P. V. Schwartz, E. J. Prinz, H. Manoharan, J. Vac. Sci. Tech., B9, 2011(1991)[4] M. K. Sanganeria, M. C. Öztürk, K. Violette, G. Harris, C. A. Lee and D. Maher,Appl. Phys. Lett. 66, 1255-57 (1995)[5] C. W. Trucks, K. Raghavachari, G. S. Higashi, Y. J. Chabal, Phys. Rev. Lett., 65,504-507 (1990)[6] B. Meyerson, F. Himpsel, K. J. Uran, Appl. Phys. Lett. 57, 1034 (1990)[7] M. Nakamori and N. Aoto, Jpn. J. Appl. Phys. pt 1, 37, 426-465 (1989)[8] "Nanochem 3000" purchased from Semi-Gas Systems, Inc. 625 Wool Creek Drive,San Jose CA 95112, 408-971-6500[9] A. St. Amour, J. C. Sturm, J. of Material Science: Materials in Electronics, 6, 350-355 (1995)[10] Z. Matutinovic-Krstelj, E. Chanson, J. C. Sturm, J. of Electronic Materials, 24,725-730 (1995)[11] C.-L. Wang, S. Unnikrishnan, B.-Y. Kim, D.-L. Kwong, A. F. Tasch, J. of Electro-chemistry, 143, 2387 (1996)[12] M. Morita, T. Ohmi, E. Hasegawa, M. Kawakami, K. Suma, Appl. Phys. Lett., 55562 (1989)[13] M. Weldon, B. Stefanov, K. Raghavachari, Y. J. Chabal, Phys. Rev. Let., 79, 2851-54 (1997)[14] V. Thanh, D. Bouchier, D. Debarre, Phys. Rev. B, 56, 10505 (1997)[15] Dumas, Y. J. Chabal, P. Jakob, Surf. Sci., 269/270, 867-78 (1992)[16] Y. Morita, H. Tokumoto, Appl. Phys. Lett., 67, 2654-56, (1995)[17] D. Gräf, M. Grundner, R. Schulz, L. Mülhoff, J. Appl. Phys., 68, 5155-61 (1990)[18] J. Westermann, H. Nienhaus, W. Mönch, Surf. Sci., 311, 101-106 (1994)[19] A. Stockhausen, T. U. Kampen, W. Mönch, Appl. Surf. Sci., 56-58, 795-801(1992)[20] T. Takahagi, A. Ishitani, H. Kuroda, Y. Nagasawa, J. Appl. Phys., 69, 803-7(1991)[20] A. Miyauchi, Y. Inoue, T. Suzuki, M. Akiyama, Appl. Phys. Lett., 57, 676-677(1990)

193

[22] K. Kinoshita, I. Nishiyama, J. Vac. Sci. Tech A, 13, 2709, (1995)[23] S. L. Cohen, J. Blum, C. D’Emic, M. Gilbert, F. Cardone, C. Stanis, M. Liehr inChemical Surface Preparation, Passivation, and Cleaning for Semiconductor Growthand Processing, R. J. Nemanich, C. R. Helms, M. Hirose, G. W. Rubloff. Editor. p.499-504. Material Research Symposium Proceedings. Pittsburgh PA (1992)[24] J. E. Northrup, Phys. Rev. B, 44,1419-1422 (1991)[25] P. M. Garone, J. C. Sturm, P. V. Schwartz, S. A. Scwartz, B. J. Wikens, Appl.Phys. Lett., 56, 1275 (1990)[26] C. C. Cheng, S. R. Lucas, H. Gutleben, W. J. Choyoke, J. T. Yates, Jr., Surf. Sci.Lett., 273, L441-L448 (1992)[27] C. C. Cheng, P. A. Taylor, R. M. Wallace, H. Gutleben, L. Clemen, M. L. Cola-ianni, P. J. Chen, W. H. Weinberg, W. J. Choyoke, J. T. Yates, Jr., Thin Solid Films,225, 196 202 (1993)[28] A. St. Amour, J. C. Sturm, Y. Lacroix, M. L. W. Thewalt, Appl. Phys. Lett., 65,3344-46 (1994)[29] K. Oda and Y. Kiyota, J. Electrochemical Society, 143, 2361 (1996)[30] J. J. Lander and J. Morrison, J. Appl. Phys., 33, 1300 (1982)[31] F. W. Smith and G. Ghidini, J. Electrochem. Soc., 128, 1300, (1982)[32] P. V. Schwartz, J. C. Sturm, J. Electrochem. Soc., 141, 1284-1290 (1994).[33] M. Yang, M. Carroll, J. C. Sturm and T. Buyuklimanli, J. Electrochem. Soc., 1473541 (2000)[34] M. Yang, C.-L. Chang, M. Carroll, J. C. Sturm, IEEE Elec. Dev. Lett., 20, 301(1999)

194

Chapter 8

Conclusion

8.1 Summary

This work was motivated by the need for ultra-sharp dopant profiles for high perfor-

mance devices. Ultra-sharp dopant profile can be obtained by techniques like ion-

implantation or epitaxy, making the primary challenge to controlling a dopant’s posi-

tion in a device its solid-state thermal diffusion during high temperature steps of the

device fabrication. Diffusion lengths as small as 1-2 nm are shown to reduce the elec-

trical performance of the SiGe HBT in chapter 3, demonstrating the sensitivity of

modern devices to very small amounts of diffusion. Two approaches to control dopant

diffusion are examined in this work: (1) reduce the dopant diffusivity through incorpo-

ration of substitutional carbon; and (2) reduce the thermal budget of critical processing

steps. To study the effect of substitutional carbon on dopant diffusivity for part of this

thesis work, it became necessary to develop a new gas chemistry (disilane) so that high

substitutional carbon concentrations (up to ~ 0.5% carbon) could be incorporated into

pure Si and low germanium concentration SiGe (7%), which is discussed in chapter 2.

Most of this work was dedicated to understanding the mechanism by which substitu-

tional carbon reduces the diffusivity of boron and phosphorus. The primary findings

were that: 0.5% substitutional carbon incorporation in SiGeC reduces the boron diffu-

sivity as much as 8 times below the intrinsic boron diffusivity; substitutional carbon

“sinks” excess interstitials introduced by ion-implantation or oxidation through a

195

direct reaction between the interstitial and the substitutional carbon; the carbon-inter-

stitial reaction may be used to non-locally suppress the enhanced diffusion effects of

device processing steps; the carbon-interstitial reaction is “kick-out” like meaning that

the reaction is one-to-one and produces mobile carbon. To quantify the carbon-inter-

stitial reaction it was furthermore necessary to quantify the interstitial injection rate

during oxidation, which included finding that the excess interstitial concentration is

fixed at the surface of the silicon-oxide interface during oxidation. Finally, a low tem-

perature cleaning technique for Si and SiGe epitaxy was developed as an alternative

method to reduce dopant diffusion for device structures that require a second growth

step.

8.2 Future Work

A primary motivation of this work was the control of dopant diffusion for sharp dopant

profiles, which can be facilitated by the local or non-local effect of carbon on the inter-

stitial concentration. However, reliable implementation of substitutional carbon in sil-

icon based devices will demand a better understanding of how to avoid conditions that

initiate the formation of incoherent silicon-carbide precipitation or carbon clusters that

are potentially electrically active.

Another technological concern is the development of a fully predictive understand-

ing of carbon and germanium’s effect on boron diffusion to reduce the costs of experi-

mental trial and error methods of optimization of a transistor structure like the SiGe(C)

196

HBT. The effect of Ge incorporation on the boron diffusivity is still not entirely under-

stood. Preliminary studies of the boron diffusivity in SiGe show that boron diffuses

more slowly with increasing germanium concentration below 40% [1], Fig 8.1. The

reduced boron diffusivity in SiGe is not believed to be due to a non-equilibrium under-

saturation of silicon self-interstitials as is proposed for SiGeC, since SiGe does not

“sink” interstitials as SiGeC does. Alternative explanations have been proposed for

the germanium effect on diffusivity including local strain trapping [2] and band-gap

effects on the defect level and carrier concentration [3-5]. However, none of these

models accounts for the ultimate increase in diffusivity that the boron must have to

reach the intrinsic boron diffusivity in pure germanium (Fig. 8.1) [6, 7]. An equally

important challenge is developing a predictive model for the effect of the SiGe/Si

transition on boron diffusion, e.g. segregation effects. Several models have been pro-

posed [5, 8-10], however these remain relatively untested and there exist almost no

reports of the effects of carbon on these properties. Therefore, significant work still

remains to be done to understand the boron diffusivity in SiGe.

197

.

Figure 8.1. Boron diffusivity dependence on germanium content in strained <100>SiGe pseudomorphic to Si compared to the boron diffusivity in pure unstrained germa-nium [1,3,11] and pure silicon.

Finally two other critical questions from a technological stand-point that remain

unresolved is the dependence of the intrinsic carbon diffusion and the rate of immobile

carbon formation, e.g. precipitation, on germanium content. Despite these concerns

and uncertainties, substitutional carbon incorporation used to control dopant diffusion

remains an active and promising approach that is not yet fully explored. The technique

has been successfully implemented in research device structures without detrimental

electrical effects [12-14] and has been successful enough to now be developed for pro-

duction in the case of the SiGeC HBT by companies like Motorola [15].

10-18

10-17

10-16

10-15

0 0.2 0.4 0.6 0.8 1

[B] Kuo et al.[B] Moriya et al.[B] Dunlap

Dif

fusi

vity

[cm

2 s-1]

Ge Content

T = 810°C

198

199

8.3 References

[1] P. Kuo, J. L. Hoyt, J. F. Gibbons, J. E. Turner, and D. Lefforge, “Boron Diffu-sion in Si and SiGe,” presented at Materials Research Symposium, 1995.[2] R. F. Lever, J. M. Bonar, and A. F. W. Willoughby, J. Appl. Phys., vol. 83, pp.1988, 1998.[3] N. Moriya, L. C. Feldman, H. S. Luftman, C. A. King, J. Bevk, and B. Freer,“Boron Diffusion in Strained SiGe Epitaxial Layers,” Phys. Rev. Lett., vol. 71, pp.883, 1993.[4] S. M. Hu, D. C. Ahlgren, P. A. Ronsheim, and J. O. Chu, “Experimental Studyof Diffusion and Segregation in a Si-(GeSi) Heterostructure,” Phys. Rev. Lett., vol. 67,pp. 1450, 1991.[5] S. M. Hu, “Diffusion and segregation in heterostructures: Theory, two limitingcases, and internal strain,” Phys. Rev. B, vol. 45, pp. 4498, 1992.[6] B. L. Sharma, presented at Defects and Diffusion Forum, 1990.[7] W. C. J. Dunlap, Phys. Rev., vol. 94, pp. 1531, 1954.[8] C.-H. Chen, U. M. Goesele, and T. Y. Tan, “Dopant diffusion and segregationin semiconductor heterostructures: Part 2. B in GexSi1-x./Si structures,” Appl. Phys. A,

vol. 68, pp. 19-24, 1999.[9] J. Tersoff, “Forces on Charged Defects in Semiconductor Heterostructures,”Phys. Rev. Lett., vol. 65, pp. 887, 1990.[10] N. Moriya, L. C. Feldman, S. W. Downeym, C. A. King, and A. B. Emerson,“Interfacial Segregation in Strained Heterostructures: Boron in Si0.8Ge0.2/Si,” Phys.

Rev. Lett., vol. 75, pp. 1981, 1995.[11] W. C. Dunlap, Phys. Rev., vol. 94, pp 1531, 1954.[12] J. H. Osten, Thin Solid Films, vol. 367, pp. 101-111, 2000.[13] M. Yang, C.-L. Chang, M. Carroll, and J. C. Sturm, “25 nm p-channel verticalMOSFETs with SiGeC source/drains,” IEEE Electr. Dev. Letters, vol. 20, pp. 301,1999.[14] H.-J. Gossmann, C. S. Rafferty, G. Hobler, H. H. Vuong, D. C. Jacobson, andM. Frei, “Suppression of reverse short channel effect by buried carbon layer,” pre-sented at Technical Digest - International Electron Device Meeting, San Fransisco,CA, USA, 1998.[15] B. Johnson, “Motorola ESTL plans to develop unique technology”,1999.

200

Appendix A

Growth Sequences

201

A.1 Low Temperature Clean Sample #2541

800oC, 1 minute, 10 torr clean used to clean a SiGe surface and grow a silicon cap onthe uncapped SiGe. Cleaning sequence is the same used for the hydrogen bakesdescribed in section 7.6.

ex-situ clean andin-situ H2 bake (800°C)

30 nm, i-Si

Si-substrate

20 nm, Si0.8Ge0.2

silicon cap subsequentto clean steps (700°C)

SiGe capped substrate

2 µm,Si (buffer layer)

202

Sequencer Table #0

Step # Action Comment

0 CONTROL ON& Turn on control

1 SCAN(0.3) Scan simulation

2 SET(SP7,0) Override power zero

3 SET(SP4,0) Turn off PID control

4 SET(SP0,0.6) Zero loop counter

5 SET(DO0,0)

6 SET(DO1,0) Hydrogen select off

7 SET(DO2,0) SiH4 select off

8 SET(DO3,0) GeH4 select off

9 SET(DO4,0) B2H6 select off

10 SET(DO5,0) PH3 select off

11 SET(DO6,0) Methylsilane (MS) select off

12 SET(DO7,0) Dichlorosilane (DCS) select off

13 SET(DO8,0) Disilane (DS) select off

14 SET(DO9,0) SiH4 inject off

15 SET(DO10,0) GeH4 inject off

16 SET(DO11,0) B2H6 inject off

17 SET(DO12,0) PH3 inject off

18 SET(DO13,0) DCS/DS inject off

19 SET(AO0,0.21) Hydrogen 1 slpm

20 SET(DO15,1) Vac on

21 SET(DO1,1) Hydrogen on

22 SET(AO1,0) SiH4 MFC zero

23 SET(AO2,0) GeH4 MFC zero

24 SET(AO3,0) B2H6 MFC zero

25 SET(AO4,0) PH3 MFC zero

26 SET(AO15,0) DS MFC zero

27 SET(AO6,0.534) DCS MFC 26 sccm

28 SET(AO7,0) MS MFC zero

29 SET(AO8,0) Pressure zero

30 SEQUENCER ON(0.3,1.0) Start sequence #1

203

Sequencer Table #1

Step # Action Comment

0 SET(SP1,1) Set layer number

1 SET(SP2, 0.0) Reset loop counter

2 WAITUNTIL(AI24>0.5)

3

4

5 SEQUENCER ON(0.3,5,0) Call clean sequence

6 WAITUNTIL(SP2>0.5) Clean sequence

7 SET(SP2,0.0) Reset loop counter

8

9 SET(AO8,0.6) Pressure 6 torr

10 WAITUNTIL(AI28<6.5) Pressure stabilize

11

12

13

14

15

16 SEQUENCER ON(0.3,3,0) Call cap sequence

17 WAITUNTIL(SP2>0.5) Cap sequence

18 SET(SP2,0.0) Reset loop counter

19

20

21

22

23

24 RAMP(SP7,-0.4,0.0) Lamps off

25

26

27

28

29 SEQUENCER ON(0.3,7,0) Call shut down

30 WAITUNTIL(SP2>0.5) Shut down

204

Sequencer Table #3

Step # Action Comment

0 SET(SP5,3.523) Set T=700

1 SET(SP4,1.0) Feed back on

2 WAIT(30) Stabilize temperature

3

4

5

6

7

8

9

10 SET(DO13,1.0) DCS inject on

11 WAIT(900) Cap layer

12 SET(D013,0) DCS inject off

13 WAIT(5) Purge tube

14

15

16

17

18

19

20

21

22

23 SET(SP4,0.0) Feedback off

24

25

26

27

28

29 SET(SP2,1.0) End

30 END

205

Sequencer Table #5

Step # Action Comment

0 SET(AO8,0.0) Pump out

1 SET(AO11,1.0) Low pressure select

2 WAITUNTIL(AI28<5.5) Stabilize pressure

3 SET(AO0,0.617) Hydrogen 3 slpm

4 SET(AO8,1.0) Set pressure 10 torr

5 WAITUNTIL(AI28>5.5) Stabilize pressure

6

7

8

9

10 WAIT(30) Stabilize pressure

11

12

13 WAITUNTIL(AI24>0.5) Go for cold values

14 SET(SP3,1) Get cold values

15 WAIT(1)

16 SET(SP3,0)

17

18

19 RAMP(SP7,0.4,0.16) Lamps on

20 WAIT(30) Warm-up wafer

21 RAMP(SP7,0.4,0.19) Approach 800 C

22 WAIT(30)

23 SET(SP5,5.488) Set T=800 C

24 SET(SP4,1.0) Feed back on

25 WAIT(60) Clean

26

27

28

29 SET(SP2,1.0) End sequence

30 END

206

Sequencer Table #7

Step # Action Comment

0 SET(SP7,0) Lamps off

1 SET(DO13,0)& DCS inject off

2 SET(DO12,0)& PH3 off

3 SET(DO11,0)& B2H6 off

4 SET(DO10,0)& SiH4 off

5 SET(DO9,0)& GeH4 off

6 SET(DO7,0)& DCS select off

7 SET(DO5,0)& PH3 off

8 SET(DO4,0)& B2H6 off

9 SET(DO3,0)& SiH4 off

10 SET(DO2,0)& GeH4 off

11 SET(DO1,0)& H2 off

12 SET(AO8,0.0)& Pump out

13 SET(AO7,0.0)& B2H6 low

14 SET(AO6,0.0)& DCS

15 SET(AO5,0.0)& PH3 low

16 SET(AO4,0.0)& PH3 high

17 SET(AO3,0.0)& B2H6 high

18 SET(AO2,0.0)& SiH4

19 SET(AO1,0.0)& GeH4

20 SET(AO0,0.00)& H2

21 WAITUNTIL(AI28<0.5) Pump out

22 SET(D015,0) Vac off

23 SEQUENCER OFF(0)

24 SEQUENCER OFF(1)

25 SEQUENCER OFF(2)

26 SEQUENCER OFF(3)

27 SEQUENCER OFF(4)

28 SEQUENCER OFF(5)

29 SEQUENCER OFF(6)

30 SEQUENCER OFF(7)

207

A.2 Si1-xCx Sample Grown w/ Disilane #2963

SiC sample grown using the disilane process described in section 2.6.3 (see Fig. 2.9for SIMS and Fig. 2.12 for XRD rocking curve).

7 nm, i-Si

Si-substrate

30 nm, Si0.996C0.004

400 nm,Si (buffer layer)

208

Sequencer Table #0

Step # Action Comment

0 CONTROL ON& Turn on control

1 SCAN(0.3) Scan simulation

2 SET(SP7,0) Override power zero

3 SET(SP4,0) Turn off PID control

4 SET(SP0,0.6) Zero loop counter

5 SET(DO0,0)

6 SET(DO1,0) Hydrogen select off

7 SET(DO2,0) SiH4 select off

8 SET(DO3,0) GeH4 select off

9 SET(DO4,0) B2H6 select off

10 SET(DO5,0) PH3 select off

11 SET(DO6,0) Methylsilane (MS) select off

12 SET(DO7,0) Dichlorosilane (DCS) select off

13 SET(DO8,0) Disilane (DS) select off

14 SET(DO9,0) SiH4 inject off

15 SET(DO10,0) GeH4 inject off

16 SET(DO11,0) B2H6 inject off

17 SET(DO12,0) PH3 inject off

18 SET(DO13,0) DCS/DS inject off

19 SET(AO0,0.21) Hydrogen 1 slpm

20 SET(DO15,1) Vac on

21 SET(DO1,1) Hydrogen on

22 SET(AO1,0) SiH4 MFC zero

23 SET(AO2,0) GeH4 MFC zero

24 SET(AO3,0) B2H6 MFC zero

25 SET(AO4,0) PH3 MFC zero

26 SET(AO15,0) DS MFC zero

27 SET(AO6,0.534) DCS MFC 26 sccm

28 SET(AO7,0) MS MFC zero

29 SET(AO8,0) Pressure zero

30 SEQUENCER ON(0.3,1.0) Start sequence #1

209

Sequencer Table #1

Step # Action Comment

0 SET(SP1,1) Set layer number

1 SET(SP2, 0.0) Reset loop counter

2 SEQUENCER ON(0.3,6,0) Call clean sequence

3 WAITUNTIL(SP2>0.5) Clean sequence

4 SET(SP2,0.0) Reset loop counter

5

6

7 SEQUENCER ON(0.3,5,0) Call clean sequence

8 WAITUNTIL(SP2>0.5) Clean sequence

9

10 SET(SP5,2.89) Set T=625

11 SET(SP4,1.0) Feedback on

12 SET(SP2,0.0) Reset loop counter

13

14

15

16 SEQUENCER ON(0.3,4,0) Call alloy layer sequence

17 WAITUNTIL(SP2>0.5) Alloy layer

18 SET(SP2,0.0) Reset loop counter

19

20

21

22

23

24 RAMP(SP7,-0.4,0.0) Lamps off

25

26

27

28

29 SEQUENCER ON(0.3,7,0) Call shut down

30 WAITUNTIL(SP2>0.5) Shut down

210

Sequencer Table #3

Step # Action Comment

0 WAIT(60)

1 SET(AO6,0.54) DCS 26 sccm

2

3 SET(DO14,1) Inject MS

4 WAIT(660) SiC layer

5 SET(DO14,0) MS Inject off

6

7 SET(DO3,0) Germane select off

8 SET(DO6,0) MS select off

9 WAIT(15) Grow Si w/ DS at 625 C

10

11

12 SET(DO8,0) DS select off

13 WAIT(10) Purge tube of DS & close hand valve

14 SET(DO7,1) Open hand valve for DCS & Select

15

16

17 SET(SP5,3.53) Set T = 700

18

19 WAIT(1020) Cap layer Si w/ DCS

20 SET(D013,0) DCS inject off

21

22

23 SET(DO7,0) DCS select off

24

25 WAIT(5) Purge

26 SET(SP5,5.5) Set T = 800 C

27 WAIT(300) Anneal layer 5 minutes

28

29 SET(SP2,1.0) End

30 END

211

Sequencer Table #5

Step # Action Comment

0 WAITUNTIL(AI24>0.5) Go for buffer layer

1 SET(AO11,1) Low P select

2 SET(AO8,0.6) Pressure 6 torr

3 WAITUNTIL(AI28>0.55) Stabilize pressure

4 SET(AO7,0.2) MS flow 0.19 sccm

5 SET(DO6,1) MS select on

6 RAMP(SP7,-0.4,0.23) T ~ 800-850 C

7 SET(DO13,1) Inject DCS

8 WAIT(600) Buffer layer

9 SET(DO13,0) Inject DCS off

10 RAMP(SP7,-0.4,0.0) Lamps off

11 SET(DO7,0) DCS select off

12 SET(AO6,0) DCS MFC off

13

14

15

16 SET(AO15,0.5) Disilane 50 sccm

17 SET(DO8,1) DS select on

18 WAITUNTIL(AI24>0.5) Go for SiC layer

19 RAMP(SP7,0.4,0.27) Lamps on

20 WAIT(5) Warm-up

21 RAMP(SP7,-0.4,0.16)

22 SET(AO8,1.0) Pressure 10 torr

23 WAIT(15) Stabilize

24

25

26 SET(DO13,1) Inject DS

27

28

29 SET(SP2,1.0) End

30 END

212

Sequencer Table #6

Step # Action Comment

0 WAITUNTIL(AI24>0.5) Go for clean

1 SET(AO11,0) High pressure select

2 SET(AO8,0.25) Pressure 250 torr

3 SET(AO0,0.817) Hydrogen 4 slpm

4 WAITUNTIL(AI29>250) Pressure stabilize

5

6

7 SET(SP3,1) Get cold values

8 WAIT(1)

9 SET(SP3,0)

10

11 SET(DO7,1) DCS select on

12 RAMP(SP7,0.4,0.27) Warm-up wafer ~900-1000 C

13 WAIT(120) Clean 2 min

14 SET(AO8,0.22) Pump out

15 WAIT(10)

16 SET(AO8,0.17)

17 WAIT(10)

18 SET(AO8,0.12)

19 WAIT(10)

20 SET(AO8,0.07)

21 WAIT(10)

22 SET(AO8,0.0)

23 SET(AO0,0.61) Hydrogen 3 slpm

24

25

26

27

28

29 SET(SP2,1.0) End

30 END

213

Appendix B

Publications and Presentations Resulting from this Thesis

Journal Articles

1. M.S. Carroll, C-L Chang, and J.C. Sturm "Complete suppression of boron tran-sient-enhanced diffusion and oxidation-enhanced diffusion in silicon usinglocalized substitutional carbon incorporation," Appl. Phys. Lett. 73, 3695-3697(1998).

2. M. Yang, C-L. Chang, M. Carroll, and J. C. Sturm, "25nm p-Channel verticalMOSFET's with SiGeC source-drains,” IEEE Elec. Dev. Lett. 20, pp. 301-303(1999).

3. M.S. Carroll, J.C. Sturm, and M. Yang, “Low-temperature preparation of oxy-gen-and carbon-free silicon-germanium surfaces for silicon and silicon-germa-nium epitaxial growth by rapid thermal chemical vapor deposition,” J.Electrochem. Soc. 147, 4652-4659 (2000).

4. M. Yang, M. Carroll, J.C. Sturm, and T. Buyuklimanli, “Phosphorus dopingand sharp profiles in silicon and silicon-germanium epitaxy by rapid thermalchemical vapor deposition,” J. Electrochem. Soc. 147, 3541-3545 (2000).

5. M. S. Carroll and J. C. Sturm, “Quantitative Measurement of the Surface Sili-con Interstitial Boundary Condition and Silicon Interstitial Injection into Sili-con during Oxidation”, Phys. Rev. B. (submitted)

6. M. S. Carroll, J. C. Sturm, E. Napolitani, D. De Salvador, M. Berti, J. Stangl,G. Bauer, D. J. Tweet, “Diffusion Enhanced Carbon Loss from SiGeC Layersdue to Oxidation”, Phys. Rev. Lett. (submitted)

214

Refereed Conference Papers

5. M. Carroll, L.L. Lanzerotti, C.C. Chang, and J.C. Sturm, "Silicon epitaxialregrowth in RTCVD for passivation of reactive ion-etched Si/SiGe/Simicrostructures," Tech. Prog. Elect. Mater. Conf., 16 (1997).

6. M. Carroll, L.L. Lanzerotti, C.C. Chang, and J.C. Sturm, "Silicon epitaxialregrowth in RTCVD for passivation of reactive ion-etched Si/SiGe/Si micro-structures," Tech. Prog. Elect. Mater. Conf., 16 (1997).

7. M.S. Carroll, L.D. Lanzerotti, and J.C. Sturm, “Quantitative measurement ofreduction of boron diffusion by substitutional carbon incorporation,” Proc.Symp. Mat. Res. Soc., 527, 417-422(1998).

8. (Invited) J.C. Sturm, M. Yang, C.L. Chang, and M.S. Carroll, “Novel Applica-tions of rapid thermal chemical vapor deposition for nanoscale MOSFET’s,”Proc. Symp. Mat. Res. Soc. 525, 273-281 (1998).

9. M.S. Carroll, C.L. Chang, J.C. Sturm, and T. Buyuklimanli, “Complete sup-pression of oxidation enhancement of boron diffusion using substitutional car-bon incorporation,” Ext. Abs. Elec. Mat. Conf, 14 (1998).

10. (Invited) J.C. Sturm., M. Yang, M.S. Carroll, and C.L. Chang, “Si1-x-yGexCy

Alloys: An Enabling Technology for Scaled High Performance Silicon-BasedHeterojunction Devices,” Proc. IEEE Silicon Monolithic Integrated Circuits inRF Systems, 1-2 (1998).

11. M. Carroll, M. Yang, and J.C. Sturm, “Ultrasharp phosphorus profiles in sili-con epitaxy by low temperature rapid thermal chemical vapor deposition,” Ext.Abs. Electrochem. Soc, p. XX (1999), and in Advances in Rapid Thermal Pro-cessing, Electrochem Soc, 99-10, pp. XX (1999).

12. (Invited) J.C. Sturm, M. Yang, and M.S. Carroll, “Doped and undoped SiGeClayers for Dopant profile control in sub-100 nm vertical MOSFET’s,” Tech.Dig. Silicon Nanoelectronics Workshop, pp. 56-57 (1999).

13. E. Stewart, M. Carroll, C.L. Chang, and J.C. Sturm, “Boron segregation inpolycrystalline Si1-x-yGexCy alloys,” Abs. Elec. Mat. Conf. pp. 18-19 (1999).

215

14. M. Yang, M.S. Carroll, and J.C. Sturm, “Doped vs. undoped SiGeC layers insub-100 nm vertical p-channel MOSFET’s,” Dig. of Int. Conf. Silicon Epitaxyand Heterostructures pp. I5-I6 (1999).

15. M.S. Carroll, J.C. Sturm, and C-L Chang, “Quantitative measurement ofreduction of phosphorus diffusion by substitutional carbon incorporation,”Proc. Symp. Mat. Res. Soc. 568, pp. 181-186, (1999).

16. M.S. Carroll, and J.C. Sturm, “Quantitative measurement of interstitial fluxand surface super-saturation during oxidation of silicon,” Proc. Symp. Mat.Res. Soc. Vol. XX, pp. XX, (2000).

17. (Invited) J.C. Sturm, M.S. Carroll, M. Yang, J. Gray, and E. Stewart, “Mecha-nisms and applications of the control of dopant profiles in silicon using Si1-x-

yGeXCy layers grown by RTCVD,” Proc. Electrochem Soc., 2000-9, pp.309-

320 (2000).

18. M.S. Carroll and J.C. Sturm, "Quantitative measurement of interstitial flux andsurface supersaturation during oxidation of silicon," Symp. Mat. Res. Soc.(April, 2000)

Conference Presentations

19. M. Carroll, L.L. Lanzerotti, C.C. Chang, and J.C. Sturm, "Silicon epitaxialregrowth in RTCVD for passivation of reactive ion-etched Si/SiGe/Si micro-structures," Electronic Materials Conference, Ft. Follins, CO, June, 1997.

20. M. Carroll, L.L. Lanzerotti, C.C. Chang, and J.C. Sturm, "Silicon epitaxialregrowth in RTCVD for passivation of reactive ion-etched Si/SiGe/Si micro-structures," Tech. Prog. Elect. Mater. Conf., Ft. Collins, CO, June, 1997.

21. M.S. Carroll, L.D. Lanzerotti, and J.C. Sturm, “Quantitative Measurement ofReduction of Boron Diffusion by Substitutional Carbon Incorporation,”.Symp. Mat. Res. Soc., San Francisco, CA (April, 1998).

22. (Invited) J.C. Sturm, M. Yang, C.L. Chang, and M.S. Carroll, “Novel Applica-tions of rapid thermal chemical vapor deposition for nanoscale MOSFET’s,”Symp. Mat. Res. Soc., San Francisco, CA (April, 1998).

216

23. M.S. Carroll, C.L. Chang, J.C. Sturm, and T. Buyuklimanli, “Complete sup-pression of oxidation enhancement of boron diffusion using substitutional car-bon incorporation,” Elec. Mat. Conf, Charlottesville, VA (June, 1998).

24. M. Yang, C.L. Chang, M. Carroll, and J.C. Sturm, “40-nm p-channel verticalMOSFET’s with SiGeC source-drains,” Device Research Conf., Charlottes-ville, VA (June, 1998).

25. (Invited) J.C. Sturm., M. Yang, M.S. Carroll, and C.L. Chang, “Si1-x-yGexCy

Alloys: An Enabling Technology for Scaled High Performance Silicon-BasedHeterojunction Devices,” Proc. IEEE Silicon Monolithic Integrated Circuits inRF Systems, Anne Arbor, MI (Sept., 1998).

26. M.S. Carroll, J.C. Sturm, and C.L. Chang, “Quantitative measurement ofreduction of phosphorus diffusion by substitutional carbon incorporation,”Symp. Mat. Res. Soc., San Francisco, CA (April, 1999).

27. M. Carroll, M. Yang, and J.C. Sturm, “Ultrasharp phosphorus profiles in sili-con epitaxy by low temperature rapid thermal chemical vapor deposition,”Symp. Electrochem. Soc., Seattle, WA (May, 1999).

28. (Invited) J.C. Sturm, M. Yang, and M.S. Carroll, “Doped and undoped SiGeClayers for Dopant profile control in sub-100 nm vertical MOSFET’s,” SiliconNanoelectronics Workshop, Osaka, Japan (June, 1999).

29. E. Stewart, M. Carroll, C.L. Chang, and J.C. Sturm, “Boron segregation inpolycrystalline Si1-x-yGexCy alloys,” Electronic. Mater. Conf., Santa Barbara,

CA (June, 1999).

30. M. Yang, M.S. Carroll, and J.C. Sturm, “Doped vs. undoped SiGeC layers insub-100 nm vertical p-channel MOSFET’s,” Int. Conf. Silicon Epitaxy andHeterostructures. Miyagi, Japan (Sept., 1999).

31. M.S. Carroll and J.C. Sturm, “Quantitative maeasurement of interstitial fluxand surface super-saturation during oxidation of silicon,” Symp. Mat. Res.Soc., San Francisco, CA (April, 2000).

32. (Invited) J.C. Sturm, M.S. Carroll, M. Yang, J. Gray, and E. Stewart, “Mecha-nisms and applications of the control of dopant profiles in silicon using Si1-x-

217

yGexCy layers grown by RTCVD,” Symp. Electochem. Soc., Toronto, Canada

(May, 2000).